Upload folder using huggingface_hub

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_173902.log

@@ -0,0 +1,1143 @@
+2023-03-04 17:39:02,644 - mmseg - INFO - Multi-processing start method is `None`
+2023-03-04 17:39:02,657 - mmseg - INFO - OpenCV num_threads is `128
+2023-03-04 17:39:02,657 - mmseg - INFO - OMP num threads is 1
+2023-03-04 17:39:02,719 - mmseg - INFO - Environment info:
+------------------------------------------------------------
+sys.platform: linux
+Python: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]
+CUDA available: True
+GPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB
+CUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch
+NVCC: Cuda compilation tools, release 11.6, V11.6.124
+GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)
+PyTorch: 1.13.1
+PyTorch compiling details: PyTorch built with:
+  - GCC 9.3
+  - C++ Version: 201402
+  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
+  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
+  - OpenMP 201511 (a.k.a. OpenMP 4.5)
+  - LAPACK is enabled (usually provided by MKL)
+  - NNPACK is enabled
+  - CPU capability usage: AVX2
+  - CUDA Runtime 11.6
+  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
+  - CuDNN 8.3.2  (built against CUDA 11.5)
+  - Magma 2.6.1
+  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 
+
+TorchVision: 0.14.1
+OpenCV: 4.7.0
+MMCV: 1.7.1
+MMCV Compiler: GCC 9.3
+MMCV CUDA Compiler: 11.6
+MMSegmentation: 0.30.0+6749699
+------------------------------------------------------------
+
+2023-03-04 17:39:02,719 - mmseg - INFO - Distributed training: True
+2023-03-04 17:39:03,384 - mmseg - INFO - Config:
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepLogits',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=166,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'
+gpu_ids = range(0, 8)
+auto_resume = True
+
+2023-03-04 17:39:07,974 - mmseg - INFO - Set random seed to 984079870, deterministic: False
+2023-03-04 17:39:08,230 - mmseg - INFO - Parameters in backbone freezed!
+2023-03-04 17:39:08,230 - mmseg - INFO - Trainable parameters in SegformerHeadUnetFCHeadSingleStep: ['unet.init_conv.weight', 'unet.init_conv.bias', 'unet.time_mlp.1.weight', 'unet.time_mlp.1.bias', 'unet.time_mlp.3.weight', 'unet.time_mlp.3.bias', 'unet.downs.0.0.mlp.1.weight', 'unet.downs.0.0.mlp.1.bias', 'unet.downs.0.0.block1.proj.weight', 'unet.downs.0.0.block1.proj.bias', 'unet.downs.0.0.block1.norm.weight', 'unet.downs.0.0.block1.norm.bias', 'unet.downs.0.0.block2.proj.weight', 'unet.downs.0.0.block2.proj.bias', 'unet.downs.0.0.block2.norm.weight', 'unet.downs.0.0.block2.norm.bias', 'unet.downs.0.1.mlp.1.weight', 'unet.downs.0.1.mlp.1.bias', 'unet.downs.0.1.block1.proj.weight', 'unet.downs.0.1.block1.proj.bias', 'unet.downs.0.1.block1.norm.weight', 'unet.downs.0.1.block1.norm.bias', 'unet.downs.0.1.block2.proj.weight', 'unet.downs.0.1.block2.proj.bias', 'unet.downs.0.1.block2.norm.weight', 'unet.downs.0.1.block2.norm.bias', 'unet.downs.0.2.fn.fn.to_qkv.weight', 'unet.downs.0.2.fn.fn.to_out.0.weight', 'unet.downs.0.2.fn.fn.to_out.0.bias', 'unet.downs.0.2.fn.fn.to_out.1.g', 'unet.downs.0.2.fn.norm.g', 'unet.downs.0.3.weight', 'unet.downs.0.3.bias', 'unet.downs.1.0.mlp.1.weight', 'unet.downs.1.0.mlp.1.bias', 'unet.downs.1.0.block1.proj.weight', 'unet.downs.1.0.block1.proj.bias', 'unet.downs.1.0.block1.norm.weight', 'unet.downs.1.0.block1.norm.bias', 'unet.downs.1.0.block2.proj.weight', 'unet.downs.1.0.block2.proj.bias', 'unet.downs.1.0.block2.norm.weight', 'unet.downs.1.0.block2.norm.bias', 'unet.downs.1.1.mlp.1.weight', 'unet.downs.1.1.mlp.1.bias', 'unet.downs.1.1.block1.proj.weight', 'unet.downs.1.1.block1.proj.bias', 'unet.downs.1.1.block1.norm.weight', 'unet.downs.1.1.block1.norm.bias', 'unet.downs.1.1.block2.proj.weight', 'unet.downs.1.1.block2.proj.bias', 'unet.downs.1.1.block2.norm.weight', 'unet.downs.1.1.block2.norm.bias', 'unet.downs.1.2.fn.fn.to_qkv.weight', 'unet.downs.1.2.fn.fn.to_out.0.weight', 'unet.downs.1.2.fn.fn.to_out.0.bias', 'unet.downs.1.2.fn.fn.to_out.1.g', 'unet.downs.1.2.fn.norm.g', 'unet.downs.1.3.weight', 'unet.downs.1.3.bias', 'unet.downs.2.0.mlp.1.weight', 'unet.downs.2.0.mlp.1.bias', 'unet.downs.2.0.block1.proj.weight', 'unet.downs.2.0.block1.proj.bias', 'unet.downs.2.0.block1.norm.weight', 'unet.downs.2.0.block1.norm.bias', 'unet.downs.2.0.block2.proj.weight', 'unet.downs.2.0.block2.proj.bias', 'unet.downs.2.0.block2.norm.weight', 'unet.downs.2.0.block2.norm.bias', 'unet.downs.2.1.mlp.1.weight', 'unet.downs.2.1.mlp.1.bias', 'unet.downs.2.1.block1.proj.weight', 'unet.downs.2.1.block1.proj.bias', 'unet.downs.2.1.block1.norm.weight', 'unet.downs.2.1.block1.norm.bias', 'unet.downs.2.1.block2.proj.weight', 'unet.downs.2.1.block2.proj.bias', 'unet.downs.2.1.block2.norm.weight', 'unet.downs.2.1.block2.norm.bias', 'unet.downs.2.2.fn.fn.to_qkv.weight', 'unet.downs.2.2.fn.fn.to_out.0.weight', 'unet.downs.2.2.fn.fn.to_out.0.bias', 'unet.downs.2.2.fn.fn.to_out.1.g', 'unet.downs.2.2.fn.norm.g', 'unet.downs.2.3.weight', 'unet.downs.2.3.bias', 'unet.ups.0.0.mlp.1.weight', 'unet.ups.0.0.mlp.1.bias', 'unet.ups.0.0.block1.proj.weight', 'unet.ups.0.0.block1.proj.bias', 'unet.ups.0.0.block1.norm.weight', 'unet.ups.0.0.block1.norm.bias', 'unet.ups.0.0.block2.proj.weight', 'unet.ups.0.0.block2.proj.bias', 'unet.ups.0.0.block2.norm.weight', 'unet.ups.0.0.block2.norm.bias', 'unet.ups.0.0.res_conv.weight', 'unet.ups.0.0.res_conv.bias', 'unet.ups.0.1.mlp.1.weight', 'unet.ups.0.1.mlp.1.bias', 'unet.ups.0.1.block1.proj.weight', 'unet.ups.0.1.block1.proj.bias', 'unet.ups.0.1.block1.norm.weight', 'unet.ups.0.1.block1.norm.bias', 'unet.ups.0.1.block2.proj.weight', 'unet.ups.0.1.block2.proj.bias', 'unet.ups.0.1.block2.norm.weight', 'unet.ups.0.1.block2.norm.bias', 'unet.ups.0.1.res_conv.weight', 'unet.ups.0.1.res_conv.bias', 'unet.ups.0.2.fn.fn.to_qkv.weight', 'unet.ups.0.2.fn.fn.to_out.0.weight', 'unet.ups.0.2.fn.fn.to_out.0.bias', 'unet.ups.0.2.fn.fn.to_out.1.g', 'unet.ups.0.2.fn.norm.g', 'unet.ups.0.3.1.weight', 'unet.ups.0.3.1.bias', 'unet.ups.1.0.mlp.1.weight', 'unet.ups.1.0.mlp.1.bias', 'unet.ups.1.0.block1.proj.weight', 'unet.ups.1.0.block1.proj.bias', 'unet.ups.1.0.block1.norm.weight', 'unet.ups.1.0.block1.norm.bias', 'unet.ups.1.0.block2.proj.weight', 'unet.ups.1.0.block2.proj.bias', 'unet.ups.1.0.block2.norm.weight', 'unet.ups.1.0.block2.norm.bias', 'unet.ups.1.0.res_conv.weight', 'unet.ups.1.0.res_conv.bias', 'unet.ups.1.1.mlp.1.weight', 'unet.ups.1.1.mlp.1.bias', 'unet.ups.1.1.block1.proj.weight', 'unet.ups.1.1.block1.proj.bias', 'unet.ups.1.1.block1.norm.weight', 'unet.ups.1.1.block1.norm.bias', 'unet.ups.1.1.block2.proj.weight', 'unet.ups.1.1.block2.proj.bias', 'unet.ups.1.1.block2.norm.weight', 'unet.ups.1.1.block2.norm.bias', 'unet.ups.1.1.res_conv.weight', 'unet.ups.1.1.res_conv.bias', 'unet.ups.1.2.fn.fn.to_qkv.weight', 'unet.ups.1.2.fn.fn.to_out.0.weight', 'unet.ups.1.2.fn.fn.to_out.0.bias', 'unet.ups.1.2.fn.fn.to_out.1.g', 'unet.ups.1.2.fn.norm.g', 'unet.ups.1.3.1.weight', 'unet.ups.1.3.1.bias', 'unet.ups.2.0.mlp.1.weight', 'unet.ups.2.0.mlp.1.bias', 'unet.ups.2.0.block1.proj.weight', 'unet.ups.2.0.block1.proj.bias', 'unet.ups.2.0.block1.norm.weight', 'unet.ups.2.0.block1.norm.bias', 'unet.ups.2.0.block2.proj.weight', 'unet.ups.2.0.block2.proj.bias', 'unet.ups.2.0.block2.norm.weight', 'unet.ups.2.0.block2.norm.bias', 'unet.ups.2.0.res_conv.weight', 'unet.ups.2.0.res_conv.bias', 'unet.ups.2.1.mlp.1.weight', 'unet.ups.2.1.mlp.1.bias', 'unet.ups.2.1.block1.proj.weight', 'unet.ups.2.1.block1.proj.bias', 'unet.ups.2.1.block1.norm.weight', 'unet.ups.2.1.block1.norm.bias', 'unet.ups.2.1.block2.proj.weight', 'unet.ups.2.1.block2.proj.bias', 'unet.ups.2.1.block2.norm.weight', 'unet.ups.2.1.block2.norm.bias', 'unet.ups.2.1.res_conv.weight', 'unet.ups.2.1.res_conv.bias', 'unet.ups.2.2.fn.fn.to_qkv.weight', 'unet.ups.2.2.fn.fn.to_out.0.weight', 'unet.ups.2.2.fn.fn.to_out.0.bias', 'unet.ups.2.2.fn.fn.to_out.1.g', 'unet.ups.2.2.fn.norm.g', 'unet.ups.2.3.weight', 'unet.ups.2.3.bias', 'unet.mid_block1.mlp.1.weight', 'unet.mid_block1.mlp.1.bias', 'unet.mid_block1.block1.proj.weight', 'unet.mid_block1.block1.proj.bias', 'unet.mid_block1.block1.norm.weight', 'unet.mid_block1.block1.norm.bias', 'unet.mid_block1.block2.proj.weight', 'unet.mid_block1.block2.proj.bias', 'unet.mid_block1.block2.norm.weight', 'unet.mid_block1.block2.norm.bias', 'unet.mid_attn.fn.fn.to_qkv.weight', 'unet.mid_attn.fn.fn.to_out.weight', 'unet.mid_attn.fn.fn.to_out.bias', 'unet.mid_attn.fn.norm.g', 'unet.mid_block2.mlp.1.weight', 'unet.mid_block2.mlp.1.bias', 'unet.mid_block2.block1.proj.weight', 'unet.mid_block2.block1.proj.bias', 'unet.mid_block2.block1.norm.weight', 'unet.mid_block2.block1.norm.bias', 'unet.mid_block2.block2.proj.weight', 'unet.mid_block2.block2.proj.bias', 'unet.mid_block2.block2.norm.weight', 'unet.mid_block2.block2.norm.bias', 'unet.final_res_block.mlp.1.weight', 'unet.final_res_block.mlp.1.bias', 'unet.final_res_block.block1.proj.weight', 'unet.final_res_block.block1.proj.bias', 'unet.final_res_block.block1.norm.weight', 'unet.final_res_block.block1.norm.bias', 'unet.final_res_block.block2.proj.weight', 'unet.final_res_block.block2.proj.bias', 'unet.final_res_block.block2.norm.weight', 'unet.final_res_block.block2.norm.bias', 'unet.final_res_block.res_conv.weight', 'unet.final_res_block.res_conv.bias', 'unet.final_conv.weight', 'unet.final_conv.bias', 'conv_seg_new.weight', 'conv_seg_new.bias']
+2023-03-04 17:39:08,231 - mmseg - INFO - Parameters in decode_head freezed!
+2023-03-04 17:39:08,250 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 17:39:08,491 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: decode_head.conv_seg.weight, decode_head.conv_seg.bias, decode_head.convs.0.conv.weight, decode_head.convs.0.bn.weight, decode_head.convs.0.bn.bias, decode_head.convs.0.bn.running_mean, decode_head.convs.0.bn.running_var, decode_head.convs.0.bn.num_batches_tracked, decode_head.convs.1.conv.weight, decode_head.convs.1.bn.weight, decode_head.convs.1.bn.bias, decode_head.convs.1.bn.running_mean, decode_head.convs.1.bn.running_var, decode_head.convs.1.bn.num_batches_tracked, decode_head.convs.2.conv.weight, decode_head.convs.2.bn.weight, decode_head.convs.2.bn.bias, decode_head.convs.2.bn.running_mean, decode_head.convs.2.bn.running_var, decode_head.convs.2.bn.num_batches_tracked, decode_head.convs.3.conv.weight, decode_head.convs.3.bn.weight, decode_head.convs.3.bn.bias, decode_head.convs.3.bn.running_mean, decode_head.convs.3.bn.running_var, decode_head.convs.3.bn.num_batches_tracked, decode_head.fusion_conv.conv.weight, decode_head.fusion_conv.bn.weight, decode_head.fusion_conv.bn.bias, decode_head.fusion_conv.bn.running_mean, decode_head.fusion_conv.bn.running_var, decode_head.fusion_conv.bn.num_batches_tracked
+
+2023-03-04 17:39:08,504 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 17:39:08,721 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: backbone.layers.0.0.projection.weight, backbone.layers.0.0.projection.bias, backbone.layers.0.0.norm.weight, backbone.layers.0.0.norm.bias, backbone.layers.0.1.0.norm1.weight, backbone.layers.0.1.0.norm1.bias, backbone.layers.0.1.0.attn.attn.in_proj_weight, backbone.layers.0.1.0.attn.attn.in_proj_bias, backbone.layers.0.1.0.attn.attn.out_proj.weight, backbone.layers.0.1.0.attn.attn.out_proj.bias, backbone.layers.0.1.0.attn.sr.weight, backbone.layers.0.1.0.attn.sr.bias, backbone.layers.0.1.0.attn.norm.weight, backbone.layers.0.1.0.attn.norm.bias, backbone.layers.0.1.0.norm2.weight, backbone.layers.0.1.0.norm2.bias, backbone.layers.0.1.0.ffn.layers.0.weight, backbone.layers.0.1.0.ffn.layers.0.bias, backbone.layers.0.1.0.ffn.layers.1.weight, backbone.layers.0.1.0.ffn.layers.1.bias, backbone.layers.0.1.0.ffn.layers.4.weight, backbone.layers.0.1.0.ffn.layers.4.bias, backbone.layers.0.1.1.norm1.weight, backbone.layers.0.1.1.norm1.bias, backbone.layers.0.1.1.attn.attn.in_proj_weight, backbone.layers.0.1.1.attn.attn.in_proj_bias, backbone.layers.0.1.1.attn.attn.out_proj.weight, backbone.layers.0.1.1.attn.attn.out_proj.bias, backbone.layers.0.1.1.attn.sr.weight, backbone.layers.0.1.1.attn.sr.bias, backbone.layers.0.1.1.attn.norm.weight, backbone.layers.0.1.1.attn.norm.bias, backbone.layers.0.1.1.norm2.weight, backbone.layers.0.1.1.norm2.bias, backbone.layers.0.1.1.ffn.layers.0.weight, backbone.layers.0.1.1.ffn.layers.0.bias, backbone.layers.0.1.1.ffn.layers.1.weight, backbone.layers.0.1.1.ffn.layers.1.bias, backbone.layers.0.1.1.ffn.layers.4.weight, backbone.layers.0.1.1.ffn.layers.4.bias, backbone.layers.0.1.2.norm1.weight, backbone.layers.0.1.2.norm1.bias, backbone.layers.0.1.2.attn.attn.in_proj_weight, backbone.layers.0.1.2.attn.attn.in_proj_bias, backbone.layers.0.1.2.attn.attn.out_proj.weight, backbone.layers.0.1.2.attn.attn.out_proj.bias, backbone.layers.0.1.2.attn.sr.weight, backbone.layers.0.1.2.attn.sr.bias, backbone.layers.0.1.2.attn.norm.weight, backbone.layers.0.1.2.attn.norm.bias, backbone.layers.0.1.2.norm2.weight, backbone.layers.0.1.2.norm2.bias, backbone.layers.0.1.2.ffn.layers.0.weight, backbone.layers.0.1.2.ffn.layers.0.bias, backbone.layers.0.1.2.ffn.layers.1.weight, backbone.layers.0.1.2.ffn.layers.1.bias, backbone.layers.0.1.2.ffn.layers.4.weight, backbone.layers.0.1.2.ffn.layers.4.bias, backbone.layers.0.2.weight, backbone.layers.0.2.bias, backbone.layers.1.0.projection.weight, backbone.layers.1.0.projection.bias, backbone.layers.1.0.norm.weight, backbone.layers.1.0.norm.bias, backbone.layers.1.1.0.norm1.weight, backbone.layers.1.1.0.norm1.bias, backbone.layers.1.1.0.attn.attn.in_proj_weight, backbone.layers.1.1.0.attn.attn.in_proj_bias, backbone.layers.1.1.0.attn.attn.out_proj.weight, backbone.layers.1.1.0.attn.attn.out_proj.bias, backbone.layers.1.1.0.attn.sr.weight, backbone.layers.1.1.0.attn.sr.bias, backbone.layers.1.1.0.attn.norm.weight, backbone.layers.1.1.0.attn.norm.bias, backbone.layers.1.1.0.norm2.weight, backbone.layers.1.1.0.norm2.bias, backbone.layers.1.1.0.ffn.layers.0.weight, backbone.layers.1.1.0.ffn.layers.0.bias, backbone.layers.1.1.0.ffn.layers.1.weight, backbone.layers.1.1.0.ffn.layers.1.bias, backbone.layers.1.1.0.ffn.layers.4.weight, backbone.layers.1.1.0.ffn.layers.4.bias, backbone.layers.1.1.1.norm1.weight, backbone.layers.1.1.1.norm1.bias, backbone.layers.1.1.1.attn.attn.in_proj_weight, backbone.layers.1.1.1.attn.attn.in_proj_bias, backbone.layers.1.1.1.attn.attn.out_proj.weight, backbone.layers.1.1.1.attn.attn.out_proj.bias, backbone.layers.1.1.1.attn.sr.weight, backbone.layers.1.1.1.attn.sr.bias, backbone.layers.1.1.1.attn.norm.weight, backbone.layers.1.1.1.attn.norm.bias, backbone.layers.1.1.1.norm2.weight, backbone.layers.1.1.1.norm2.bias, backbone.layers.1.1.1.ffn.layers.0.weight, backbone.layers.1.1.1.ffn.layers.0.bias, backbone.layers.1.1.1.ffn.layers.1.weight, backbone.layers.1.1.1.ffn.layers.1.bias, backbone.layers.1.1.1.ffn.layers.4.weight, backbone.layers.1.1.1.ffn.layers.4.bias, backbone.layers.1.1.2.norm1.weight, backbone.layers.1.1.2.norm1.bias, backbone.layers.1.1.2.attn.attn.in_proj_weight, backbone.layers.1.1.2.attn.attn.in_proj_bias, backbone.layers.1.1.2.attn.attn.out_proj.weight, backbone.layers.1.1.2.attn.attn.out_proj.bias, backbone.layers.1.1.2.attn.sr.weight, backbone.layers.1.1.2.attn.sr.bias, backbone.layers.1.1.2.attn.norm.weight, backbone.layers.1.1.2.attn.norm.bias, backbone.layers.1.1.2.norm2.weight, backbone.layers.1.1.2.norm2.bias, backbone.layers.1.1.2.ffn.layers.0.weight, backbone.layers.1.1.2.ffn.layers.0.bias, backbone.layers.1.1.2.ffn.layers.1.weight, backbone.layers.1.1.2.ffn.layers.1.bias, backbone.layers.1.1.2.ffn.layers.4.weight, backbone.layers.1.1.2.ffn.layers.4.bias, backbone.layers.1.1.3.norm1.weight, backbone.layers.1.1.3.norm1.bias, backbone.layers.1.1.3.attn.attn.in_proj_weight, backbone.layers.1.1.3.attn.attn.in_proj_bias, backbone.layers.1.1.3.attn.attn.out_proj.weight, backbone.layers.1.1.3.attn.attn.out_proj.bias, backbone.layers.1.1.3.attn.sr.weight, backbone.layers.1.1.3.attn.sr.bias, backbone.layers.1.1.3.attn.norm.weight, backbone.layers.1.1.3.attn.norm.bias, backbone.layers.1.1.3.norm2.weight, backbone.layers.1.1.3.norm2.bias, backbone.layers.1.1.3.ffn.layers.0.weight, backbone.layers.1.1.3.ffn.layers.0.bias, backbone.layers.1.1.3.ffn.layers.1.weight, backbone.layers.1.1.3.ffn.layers.1.bias, backbone.layers.1.1.3.ffn.layers.4.weight, backbone.layers.1.1.3.ffn.layers.4.bias, backbone.layers.1.2.weight, backbone.layers.1.2.bias, backbone.layers.2.0.projection.weight, backbone.layers.2.0.projection.bias, backbone.layers.2.0.norm.weight, backbone.layers.2.0.norm.bias, backbone.layers.2.1.0.norm1.weight, backbone.layers.2.1.0.norm1.bias, backbone.layers.2.1.0.attn.attn.in_proj_weight, backbone.layers.2.1.0.attn.attn.in_proj_bias, backbone.layers.2.1.0.attn.attn.out_proj.weight, backbone.layers.2.1.0.attn.attn.out_proj.bias, backbone.layers.2.1.0.attn.sr.weight, backbone.layers.2.1.0.attn.sr.bias, backbone.layers.2.1.0.attn.norm.weight, backbone.layers.2.1.0.attn.norm.bias, backbone.layers.2.1.0.norm2.weight, backbone.layers.2.1.0.norm2.bias, backbone.layers.2.1.0.ffn.layers.0.weight, backbone.layers.2.1.0.ffn.layers.0.bias, backbone.layers.2.1.0.ffn.layers.1.weight, backbone.layers.2.1.0.ffn.layers.1.bias, backbone.layers.2.1.0.ffn.layers.4.weight, backbone.layers.2.1.0.ffn.layers.4.bias, backbone.layers.2.1.1.norm1.weight, backbone.layers.2.1.1.norm1.bias, backbone.layers.2.1.1.attn.attn.in_proj_weight, backbone.layers.2.1.1.attn.attn.in_proj_bias, backbone.layers.2.1.1.attn.attn.out_proj.weight, backbone.layers.2.1.1.attn.attn.out_proj.bias, backbone.layers.2.1.1.attn.sr.weight, backbone.layers.2.1.1.attn.sr.bias, backbone.layers.2.1.1.attn.norm.weight, backbone.layers.2.1.1.attn.norm.bias, backbone.layers.2.1.1.norm2.weight, backbone.layers.2.1.1.norm2.bias, backbone.layers.2.1.1.ffn.layers.0.weight, backbone.layers.2.1.1.ffn.layers.0.bias, backbone.layers.2.1.1.ffn.layers.1.weight, backbone.layers.2.1.1.ffn.layers.1.bias, backbone.layers.2.1.1.ffn.layers.4.weight, backbone.layers.2.1.1.ffn.layers.4.bias, backbone.layers.2.1.2.norm1.weight, backbone.layers.2.1.2.norm1.bias, backbone.layers.2.1.2.attn.attn.in_proj_weight, backbone.layers.2.1.2.attn.attn.in_proj_bias, backbone.layers.2.1.2.attn.attn.out_proj.weight, backbone.layers.2.1.2.attn.attn.out_proj.bias, backbone.layers.2.1.2.attn.sr.weight, backbone.layers.2.1.2.attn.sr.bias, backbone.layers.2.1.2.attn.norm.weight, backbone.layers.2.1.2.attn.norm.bias, backbone.layers.2.1.2.norm2.weight, backbone.layers.2.1.2.norm2.bias, backbone.layers.2.1.2.ffn.layers.0.weight, backbone.layers.2.1.2.ffn.layers.0.bias, backbone.layers.2.1.2.ffn.layers.1.weight, backbone.layers.2.1.2.ffn.layers.1.bias, backbone.layers.2.1.2.ffn.layers.4.weight, backbone.layers.2.1.2.ffn.layers.4.bias, backbone.layers.2.1.3.norm1.weight, backbone.layers.2.1.3.norm1.bias, backbone.layers.2.1.3.attn.attn.in_proj_weight, backbone.layers.2.1.3.attn.attn.in_proj_bias, backbone.layers.2.1.3.attn.attn.out_proj.weight, backbone.layers.2.1.3.attn.attn.out_proj.bias, backbone.layers.2.1.3.attn.sr.weight, backbone.layers.2.1.3.attn.sr.bias, backbone.layers.2.1.3.attn.norm.weight, backbone.layers.2.1.3.attn.norm.bias, backbone.layers.2.1.3.norm2.weight, backbone.layers.2.1.3.norm2.bias, backbone.layers.2.1.3.ffn.layers.0.weight, backbone.layers.2.1.3.ffn.layers.0.bias, backbone.layers.2.1.3.ffn.layers.1.weight, backbone.layers.2.1.3.ffn.layers.1.bias, backbone.layers.2.1.3.ffn.layers.4.weight, backbone.layers.2.1.3.ffn.layers.4.bias, backbone.layers.2.1.4.norm1.weight, backbone.layers.2.1.4.norm1.bias, backbone.layers.2.1.4.attn.attn.in_proj_weight, backbone.layers.2.1.4.attn.attn.in_proj_bias, backbone.layers.2.1.4.attn.attn.out_proj.weight, backbone.layers.2.1.4.attn.attn.out_proj.bias, backbone.layers.2.1.4.attn.sr.weight, backbone.layers.2.1.4.attn.sr.bias, backbone.layers.2.1.4.attn.norm.weight, backbone.layers.2.1.4.attn.norm.bias, backbone.layers.2.1.4.norm2.weight, backbone.layers.2.1.4.norm2.bias, backbone.layers.2.1.4.ffn.layers.0.weight, backbone.layers.2.1.4.ffn.layers.0.bias, backbone.layers.2.1.4.ffn.layers.1.weight, backbone.layers.2.1.4.ffn.layers.1.bias, backbone.layers.2.1.4.ffn.layers.4.weight, backbone.layers.2.1.4.ffn.layers.4.bias, backbone.layers.2.1.5.norm1.weight, backbone.layers.2.1.5.norm1.bias, backbone.layers.2.1.5.attn.attn.in_proj_weight, backbone.layers.2.1.5.attn.attn.in_proj_bias, backbone.layers.2.1.5.attn.attn.out_proj.weight, backbone.layers.2.1.5.attn.attn.out_proj.bias, backbone.layers.2.1.5.attn.sr.weight, backbone.layers.2.1.5.attn.sr.bias, backbone.layers.2.1.5.attn.norm.weight, backbone.layers.2.1.5.attn.norm.bias, backbone.layers.2.1.5.norm2.weight, backbone.layers.2.1.5.norm2.bias, backbone.layers.2.1.5.ffn.layers.0.weight, backbone.layers.2.1.5.ffn.layers.0.bias, backbone.layers.2.1.5.ffn.layers.1.weight, backbone.layers.2.1.5.ffn.layers.1.bias, backbone.layers.2.1.5.ffn.layers.4.weight, backbone.layers.2.1.5.ffn.layers.4.bias, backbone.layers.2.2.weight, backbone.layers.2.2.bias, backbone.layers.3.0.projection.weight, backbone.layers.3.0.projection.bias, backbone.layers.3.0.norm.weight, backbone.layers.3.0.norm.bias, backbone.layers.3.1.0.norm1.weight, backbone.layers.3.1.0.norm1.bias, backbone.layers.3.1.0.attn.attn.in_proj_weight, backbone.layers.3.1.0.attn.attn.in_proj_bias, backbone.layers.3.1.0.attn.attn.out_proj.weight, backbone.layers.3.1.0.attn.attn.out_proj.bias, backbone.layers.3.1.0.norm2.weight, backbone.layers.3.1.0.norm2.bias, backbone.layers.3.1.0.ffn.layers.0.weight, backbone.layers.3.1.0.ffn.layers.0.bias, backbone.layers.3.1.0.ffn.layers.1.weight, backbone.layers.3.1.0.ffn.layers.1.bias, backbone.layers.3.1.0.ffn.layers.4.weight, backbone.layers.3.1.0.ffn.layers.4.bias, backbone.layers.3.1.1.norm1.weight, backbone.layers.3.1.1.norm1.bias, backbone.layers.3.1.1.attn.attn.in_proj_weight, backbone.layers.3.1.1.attn.attn.in_proj_bias, backbone.layers.3.1.1.attn.attn.out_proj.weight, backbone.layers.3.1.1.attn.attn.out_proj.bias, backbone.layers.3.1.1.norm2.weight, backbone.layers.3.1.1.norm2.bias, backbone.layers.3.1.1.ffn.layers.0.weight, backbone.layers.3.1.1.ffn.layers.0.bias, backbone.layers.3.1.1.ffn.layers.1.weight, backbone.layers.3.1.1.ffn.layers.1.bias, backbone.layers.3.1.1.ffn.layers.4.weight, backbone.layers.3.1.1.ffn.layers.4.bias, backbone.layers.3.1.2.norm1.weight, backbone.layers.3.1.2.norm1.bias, backbone.layers.3.1.2.attn.attn.in_proj_weight, backbone.layers.3.1.2.attn.attn.in_proj_bias, backbone.layers.3.1.2.attn.attn.out_proj.weight, backbone.layers.3.1.2.attn.attn.out_proj.bias, backbone.layers.3.1.2.norm2.weight, backbone.layers.3.1.2.norm2.bias, backbone.layers.3.1.2.ffn.layers.0.weight, backbone.layers.3.1.2.ffn.layers.0.bias, backbone.layers.3.1.2.ffn.layers.1.weight, backbone.layers.3.1.2.ffn.layers.1.bias, backbone.layers.3.1.2.ffn.layers.4.weight, backbone.layers.3.1.2.ffn.layers.4.bias, backbone.layers.3.2.weight, backbone.layers.3.2.bias
+
+missing keys in source state_dict: unet.init_conv.weight, unet.init_conv.bias, unet.time_mlp.1.weight, unet.time_mlp.1.bias, unet.time_mlp.3.weight, unet.time_mlp.3.bias, unet.downs.0.0.mlp.1.weight, unet.downs.0.0.mlp.1.bias, unet.downs.0.0.block1.proj.weight, unet.downs.0.0.block1.proj.bias, unet.downs.0.0.block1.norm.weight, unet.downs.0.0.block1.norm.bias, unet.downs.0.0.block2.proj.weight, unet.downs.0.0.block2.proj.bias, unet.downs.0.0.block2.norm.weight, unet.downs.0.0.block2.norm.bias, unet.downs.0.1.mlp.1.weight, unet.downs.0.1.mlp.1.bias, unet.downs.0.1.block1.proj.weight, unet.downs.0.1.block1.proj.bias, unet.downs.0.1.block1.norm.weight, unet.downs.0.1.block1.norm.bias, unet.downs.0.1.block2.proj.weight, unet.downs.0.1.block2.proj.bias, unet.downs.0.1.block2.norm.weight, unet.downs.0.1.block2.norm.bias, unet.downs.0.2.fn.fn.to_qkv.weight, unet.downs.0.2.fn.fn.to_out.0.weight, unet.downs.0.2.fn.fn.to_out.0.bias, unet.downs.0.2.fn.fn.to_out.1.g, unet.downs.0.2.fn.norm.g, unet.downs.0.3.weight, unet.downs.0.3.bias, unet.downs.1.0.mlp.1.weight, unet.downs.1.0.mlp.1.bias, unet.downs.1.0.block1.proj.weight, unet.downs.1.0.block1.proj.bias, unet.downs.1.0.block1.norm.weight, unet.downs.1.0.block1.norm.bias, unet.downs.1.0.block2.proj.weight, unet.downs.1.0.block2.proj.bias, unet.downs.1.0.block2.norm.weight, unet.downs.1.0.block2.norm.bias, unet.downs.1.1.mlp.1.weight, unet.downs.1.1.mlp.1.bias, unet.downs.1.1.block1.proj.weight, unet.downs.1.1.block1.proj.bias, unet.downs.1.1.block1.norm.weight, unet.downs.1.1.block1.norm.bias, unet.downs.1.1.block2.proj.weight, unet.downs.1.1.block2.proj.bias, unet.downs.1.1.block2.norm.weight, unet.downs.1.1.block2.norm.bias, unet.downs.1.2.fn.fn.to_qkv.weight, unet.downs.1.2.fn.fn.to_out.0.weight, unet.downs.1.2.fn.fn.to_out.0.bias, unet.downs.1.2.fn.fn.to_out.1.g, unet.downs.1.2.fn.norm.g, unet.downs.1.3.weight, unet.downs.1.3.bias, unet.downs.2.0.mlp.1.weight, unet.downs.2.0.mlp.1.bias, unet.downs.2.0.block1.proj.weight, unet.downs.2.0.block1.proj.bias, unet.downs.2.0.block1.norm.weight, unet.downs.2.0.block1.norm.bias, unet.downs.2.0.block2.proj.weight, unet.downs.2.0.block2.proj.bias, unet.downs.2.0.block2.norm.weight, unet.downs.2.0.block2.norm.bias, unet.downs.2.1.mlp.1.weight, unet.downs.2.1.mlp.1.bias, unet.downs.2.1.block1.proj.weight, unet.downs.2.1.block1.proj.bias, unet.downs.2.1.block1.norm.weight, unet.downs.2.1.block1.norm.bias, unet.downs.2.1.block2.proj.weight, unet.downs.2.1.block2.proj.bias, unet.downs.2.1.block2.norm.weight, unet.downs.2.1.block2.norm.bias, unet.downs.2.2.fn.fn.to_qkv.weight, unet.downs.2.2.fn.fn.to_out.0.weight, unet.downs.2.2.fn.fn.to_out.0.bias, unet.downs.2.2.fn.fn.to_out.1.g, unet.downs.2.2.fn.norm.g, unet.downs.2.3.weight, unet.downs.2.3.bias, unet.ups.0.0.mlp.1.weight, unet.ups.0.0.mlp.1.bias, unet.ups.0.0.block1.proj.weight, unet.ups.0.0.block1.proj.bias, unet.ups.0.0.block1.norm.weight, unet.ups.0.0.block1.norm.bias, unet.ups.0.0.block2.proj.weight, unet.ups.0.0.block2.proj.bias, unet.ups.0.0.block2.norm.weight, unet.ups.0.0.block2.norm.bias, unet.ups.0.0.res_conv.weight, unet.ups.0.0.res_conv.bias, unet.ups.0.1.mlp.1.weight, unet.ups.0.1.mlp.1.bias, unet.ups.0.1.block1.proj.weight, unet.ups.0.1.block1.proj.bias, unet.ups.0.1.block1.norm.weight, unet.ups.0.1.block1.norm.bias, unet.ups.0.1.block2.proj.weight, unet.ups.0.1.block2.proj.bias, unet.ups.0.1.block2.norm.weight, unet.ups.0.1.block2.norm.bias, unet.ups.0.1.res_conv.weight, unet.ups.0.1.res_conv.bias, unet.ups.0.2.fn.fn.to_qkv.weight, unet.ups.0.2.fn.fn.to_out.0.weight, unet.ups.0.2.fn.fn.to_out.0.bias, unet.ups.0.2.fn.fn.to_out.1.g, unet.ups.0.2.fn.norm.g, unet.ups.0.3.1.weight, unet.ups.0.3.1.bias, unet.ups.1.0.mlp.1.weight, unet.ups.1.0.mlp.1.bias, unet.ups.1.0.block1.proj.weight, unet.ups.1.0.block1.proj.bias, unet.ups.1.0.block1.norm.weight, unet.ups.1.0.block1.norm.bias, unet.ups.1.0.block2.proj.weight, unet.ups.1.0.block2.proj.bias, unet.ups.1.0.block2.norm.weight, unet.ups.1.0.block2.norm.bias, unet.ups.1.0.res_conv.weight, unet.ups.1.0.res_conv.bias, unet.ups.1.1.mlp.1.weight, unet.ups.1.1.mlp.1.bias, unet.ups.1.1.block1.proj.weight, unet.ups.1.1.block1.proj.bias, unet.ups.1.1.block1.norm.weight, unet.ups.1.1.block1.norm.bias, unet.ups.1.1.block2.proj.weight, unet.ups.1.1.block2.proj.bias, unet.ups.1.1.block2.norm.weight, unet.ups.1.1.block2.norm.bias, unet.ups.1.1.res_conv.weight, unet.ups.1.1.res_conv.bias, unet.ups.1.2.fn.fn.to_qkv.weight, unet.ups.1.2.fn.fn.to_out.0.weight, unet.ups.1.2.fn.fn.to_out.0.bias, unet.ups.1.2.fn.fn.to_out.1.g, unet.ups.1.2.fn.norm.g, unet.ups.1.3.1.weight, unet.ups.1.3.1.bias, unet.ups.2.0.mlp.1.weight, unet.ups.2.0.mlp.1.bias, unet.ups.2.0.block1.proj.weight, unet.ups.2.0.block1.proj.bias, unet.ups.2.0.block1.norm.weight, unet.ups.2.0.block1.norm.bias, unet.ups.2.0.block2.proj.weight, unet.ups.2.0.block2.proj.bias, unet.ups.2.0.block2.norm.weight, unet.ups.2.0.block2.norm.bias, unet.ups.2.0.res_conv.weight, unet.ups.2.0.res_conv.bias, unet.ups.2.1.mlp.1.weight, unet.ups.2.1.mlp.1.bias, unet.ups.2.1.block1.proj.weight, unet.ups.2.1.block1.proj.bias, unet.ups.2.1.block1.norm.weight, unet.ups.2.1.block1.norm.bias, unet.ups.2.1.block2.proj.weight, unet.ups.2.1.block2.proj.bias, unet.ups.2.1.block2.norm.weight, unet.ups.2.1.block2.norm.bias, unet.ups.2.1.res_conv.weight, unet.ups.2.1.res_conv.bias, unet.ups.2.2.fn.fn.to_qkv.weight, unet.ups.2.2.fn.fn.to_out.0.weight, unet.ups.2.2.fn.fn.to_out.0.bias, unet.ups.2.2.fn.fn.to_out.1.g, unet.ups.2.2.fn.norm.g, unet.ups.2.3.weight, unet.ups.2.3.bias, unet.mid_block1.mlp.1.weight, unet.mid_block1.mlp.1.bias, unet.mid_block1.block1.proj.weight, unet.mid_block1.block1.proj.bias, unet.mid_block1.block1.norm.weight, unet.mid_block1.block1.norm.bias, unet.mid_block1.block2.proj.weight, unet.mid_block1.block2.proj.bias, unet.mid_block1.block2.norm.weight, unet.mid_block1.block2.norm.bias, unet.mid_attn.fn.fn.to_qkv.weight, unet.mid_attn.fn.fn.to_out.weight, unet.mid_attn.fn.fn.to_out.bias, unet.mid_attn.fn.norm.g, unet.mid_block2.mlp.1.weight, unet.mid_block2.mlp.1.bias, unet.mid_block2.block1.proj.weight, unet.mid_block2.block1.proj.bias, unet.mid_block2.block1.norm.weight, unet.mid_block2.block1.norm.bias, unet.mid_block2.block2.proj.weight, unet.mid_block2.block2.proj.bias, unet.mid_block2.block2.norm.weight, unet.mid_block2.block2.norm.bias, unet.final_res_block.mlp.1.weight, unet.final_res_block.mlp.1.bias, unet.final_res_block.block1.proj.weight, unet.final_res_block.block1.proj.bias, unet.final_res_block.block1.norm.weight, unet.final_res_block.block1.norm.bias, unet.final_res_block.block2.proj.weight, unet.final_res_block.block2.proj.bias, unet.final_res_block.block2.norm.weight, unet.final_res_block.block2.norm.bias, unet.final_res_block.res_conv.weight, unet.final_res_block.res_conv.bias, unet.final_conv.weight, unet.final_conv.bias, conv_seg_new.weight, conv_seg_new.bias, embed.weight
+
+2023-03-04 17:39:08,744 - mmseg - INFO - EncoderDecoderFreeze(
+  (backbone): MixVisionTransformerCustomInitWeights(
+    (layers): ModuleList(
+      (0): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(3, 64, kernel_size=(7, 7), stride=(4, 4), padding=(3, 3))
+          (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+      )
+      (1): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+      )
+      (2): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(128, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (4): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (5): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+      )
+      (3): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(320, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+      )
+    )
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+  (decode_head): SegformerHeadUnetFCHeadSingleStepLogits(
+    input_transform=multiple_select, ignore_index=0, align_corners=False
+    (loss_decode): CrossEntropyLoss(avg_non_ignore=False)
+    (conv_seg): Conv2d(256, 150, kernel_size=(1, 1), stride=(1, 1))
+    (dropout): Dropout2d(p=0.1, inplace=False)
+    (convs): ModuleList(
+      (0): ConvModule(
+        (conv): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (1): ConvModule(
+        (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (2): ConvModule(
+        (conv): Conv2d(320, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (3): ConvModule(
+        (conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+    )
+    (fusion_conv): ConvModule(
+      (conv): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+      (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+      (activate): ReLU(inplace=True)
+    )
+    (unet): Unet(
+      (init_conv): Conv2d(166, 128, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3))
+      (time_mlp): Sequential(
+        (0): SinusoidalPosEmb()
+        (1): Linear(in_features=128, out_features=512, bias=True)
+        (2): GELU(approximate='none')
+        (3): Linear(in_features=512, out_features=512, bias=True)
+      )
+      (downs): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (ups): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (mid_block1): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (mid_attn): Residual(
+        (fn): PreNorm(
+          (fn): Attention(
+            (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+            (to_out): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (norm): LayerNorm()
+        )
+      )
+      (mid_block2): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (final_res_block): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+      )
+      (final_conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1))
+    )
+    (conv_seg_new): Conv2d(256, 151, kernel_size=(1, 1), stride=(1, 1))
+    (embed): Embedding(151, 16)
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+)
+2023-03-04 17:39:09,635 - mmseg - INFO - Loaded 20210 images
+2023-03-04 17:39:10,639 - mmseg - INFO - Loaded 2000 images
+2023-03-04 17:39:10,642 - mmseg - INFO - Start running, host: laizeqiang@SH-IDC1-10-140-37-130, work_dir: /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits
+2023-03-04 17:39:10,642 - mmseg - INFO - Hooks will be executed in the following order:
+before_run:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_epoch:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_iter:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+ -------------------- 
+after_train_iter:
+(ABOVE_NORMAL) OptimizerHook                      
+(NORMAL      ) CheckpointHook                     
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_train_epoch:
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_epoch:
+(LOW         ) IterTimerHook                      
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_epoch:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_run:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+2023-03-04 17:39:10,642 - mmseg - INFO - workflow: [('train', 1)], max: 80000 iters
+2023-03-04 17:39:10,642 - mmseg - INFO - Checkpoints will be saved to /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits by HardDiskBackend.
+2023-03-04 17:39:48,984 - mmseg - INFO - Iter [50/80000]	lr: 7.350e-06, eta: 6:37:37, time: 0.298, data_time: 0.015, memory: 19750, decode.loss_ce: 4.0785, decode.acc_seg: 8.5126, loss: 4.0785
+2023-03-04 17:39:57,554 - mmseg - INFO - Iter [100/80000]	lr: 1.485e-05, eta: 5:12:50, time: 0.171, data_time: 0.007, memory: 19750, decode.loss_ce: 2.9187, decode.acc_seg: 27.5140, loss: 2.9187
+2023-03-04 17:40:06,040 - mmseg - INFO - Iter [150/80000]	lr: 2.235e-05, eta: 4:43:42, time: 0.170, data_time: 0.007, memory: 19750, decode.loss_ce: 2.3354, decode.acc_seg: 43.1981, loss: 2.3354
+2023-03-04 17:40:14,295 - mmseg - INFO - Iter [200/80000]	lr: 2.985e-05, eta: 4:27:32, time: 0.165, data_time: 0.007, memory: 19750, decode.loss_ce: 1.8341, decode.acc_seg: 55.2996, loss: 1.8341
+2023-03-04 17:40:22,579 - mmseg - INFO - Iter [250/80000]	lr: 3.735e-05, eta: 4:17:56, time: 0.166, data_time: 0.007, memory: 19750, decode.loss_ce: 1.5030, decode.acc_seg: 63.0600, loss: 1.5030
+2023-03-04 17:40:30,864 - mmseg - INFO - Iter [300/80000]	lr: 4.485e-05, eta: 4:11:29, time: 0.166, data_time: 0.006, memory: 19750, decode.loss_ce: 1.2782, decode.acc_seg: 67.0304, loss: 1.2782

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_173902.log.json

@@ -0,0 +1,7 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+6749699", "seed": 984079870, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits.py", "mmseg_version": "0.30.0+6749699", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepLogits',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=166,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 984079870\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}
+{"mode": "train", "epoch": 1, "iter": 50, "lr": 1e-05, "memory": 19750, "data_time": 0.01458, "decode.loss_ce": 4.07853, "decode.acc_seg": 8.51256, "loss": 4.07853, "time": 0.2984}
+{"mode": "train", "epoch": 1, "iter": 100, "lr": 1e-05, "memory": 19750, "data_time": 0.00696, "decode.loss_ce": 2.91874, "decode.acc_seg": 27.51403, "loss": 2.91874, "time": 0.17144}
+{"mode": "train", "epoch": 1, "iter": 150, "lr": 2e-05, "memory": 19750, "data_time": 0.00722, "decode.loss_ce": 2.33541, "decode.acc_seg": 43.1981, "loss": 2.33541, "time": 0.16968}
+{"mode": "train", "epoch": 1, "iter": 200, "lr": 3e-05, "memory": 19750, "data_time": 0.00706, "decode.loss_ce": 1.83407, "decode.acc_seg": 55.29959, "loss": 1.83407, "time": 0.16511}
+{"mode": "train", "epoch": 1, "iter": 250, "lr": 4e-05, "memory": 19750, "data_time": 0.00692, "decode.loss_ce": 1.50299, "decode.acc_seg": 63.05997, "loss": 1.50299, "time": 0.16567}
+{"mode": "train", "epoch": 1, "iter": 300, "lr": 4e-05, "memory": 19750, "data_time": 0.00637, "decode.loss_ce": 1.27818, "decode.acc_seg": 67.03043, "loss": 1.27818, "time": 0.16569}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_174053.log.json

@@ -0,0 +1,161 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+6749699", "seed": 358795777, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits.py", "mmseg_version": "0.30.0+6749699", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepLogits',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=166,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 358795777\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}
+{"mode": "train", "epoch": 1, "iter": 50, "lr": 1e-05, "memory": 19750, "data_time": 0.01536, "decode.loss_ce": 3.70671, "decode.acc_seg": 12.7455, "loss": 3.70671, "time": 0.29152}
+{"mode": "train", "epoch": 1, "iter": 100, "lr": 1e-05, "memory": 19750, "data_time": 0.0067, "decode.loss_ce": 2.83767, "decode.acc_seg": 33.79282, "loss": 2.83767, "time": 0.17411}
+{"mode": "train", "epoch": 1, "iter": 150, "lr": 2e-05, "memory": 19750, "data_time": 0.00681, "decode.loss_ce": 2.29515, "decode.acc_seg": 45.9183, "loss": 2.29515, "time": 0.17116}
+{"mode": "train", "epoch": 1, "iter": 200, "lr": 3e-05, "memory": 19750, "data_time": 0.00602, "decode.loss_ce": 1.87201, "decode.acc_seg": 55.43423, "loss": 1.87201, "time": 0.1687}
+{"mode": "train", "epoch": 1, "iter": 250, "lr": 4e-05, "memory": 19750, "data_time": 0.0063, "decode.loss_ce": 1.58648, "decode.acc_seg": 61.11375, "loss": 1.58648, "time": 0.16766}
+{"mode": "train", "epoch": 1, "iter": 300, "lr": 4e-05, "memory": 19750, "data_time": 0.00626, "decode.loss_ce": 1.29469, "decode.acc_seg": 67.78908, "loss": 1.29469, "time": 0.16954}
+{"mode": "train", "epoch": 1, "iter": 350, "lr": 5e-05, "memory": 19750, "data_time": 0.00687, "decode.loss_ce": 1.17611, "decode.acc_seg": 69.68662, "loss": 1.17611, "time": 0.16891}
+{"mode": "train", "epoch": 1, "iter": 400, "lr": 6e-05, "memory": 19750, "data_time": 0.00608, "decode.loss_ce": 1.04045, "decode.acc_seg": 72.46196, "loss": 1.04045, "time": 0.16652}
+{"mode": "train", "epoch": 1, "iter": 450, "lr": 7e-05, "memory": 19750, "data_time": 0.00665, "decode.loss_ce": 0.90291, "decode.acc_seg": 73.95125, "loss": 0.90291, "time": 0.16492}
+{"mode": "train", "epoch": 1, "iter": 500, "lr": 7e-05, "memory": 19750, "data_time": 0.00686, "decode.loss_ce": 0.84245, "decode.acc_seg": 75.08308, "loss": 0.84245, "time": 0.16755}
+{"mode": "train", "epoch": 1, "iter": 550, "lr": 8e-05, "memory": 19750, "data_time": 0.00668, "decode.loss_ce": 0.72894, "decode.acc_seg": 77.32032, "loss": 0.72894, "time": 0.17103}
+{"mode": "train", "epoch": 1, "iter": 600, "lr": 9e-05, "memory": 19750, "data_time": 0.00692, "decode.loss_ce": 0.70877, "decode.acc_seg": 77.56479, "loss": 0.70877, "time": 0.16587}
+{"mode": "train", "epoch": 2, "iter": 650, "lr": 0.0001, "memory": 19750, "data_time": 0.05444, "decode.loss_ce": 0.70565, "decode.acc_seg": 77.28747, "loss": 0.70565, "time": 0.21387}
+{"mode": "train", "epoch": 2, "iter": 700, "lr": 0.0001, "memory": 19750, "data_time": 0.00651, "decode.loss_ce": 0.60632, "decode.acc_seg": 79.73918, "loss": 0.60632, "time": 0.16371}
+{"mode": "train", "epoch": 2, "iter": 750, "lr": 0.00011, "memory": 19750, "data_time": 0.00707, "decode.loss_ce": 0.59834, "decode.acc_seg": 79.62232, "loss": 0.59834, "time": 0.16193}
+{"mode": "train", "epoch": 2, "iter": 800, "lr": 0.00012, "memory": 19750, "data_time": 0.00665, "decode.loss_ce": 0.60848, "decode.acc_seg": 79.37198, "loss": 0.60848, "time": 0.16804}
+{"mode": "train", "epoch": 2, "iter": 850, "lr": 0.00013, "memory": 19750, "data_time": 0.0067, "decode.loss_ce": 0.55236, "decode.acc_seg": 80.7622, "loss": 0.55236, "time": 0.17566}
+{"mode": "train", "epoch": 2, "iter": 900, "lr": 0.00013, "memory": 19750, "data_time": 0.00684, "decode.loss_ce": 0.51953, "decode.acc_seg": 81.87914, "loss": 0.51953, "time": 0.17055}
+{"mode": "train", "epoch": 2, "iter": 950, "lr": 0.00014, "memory": 19750, "data_time": 0.00679, "decode.loss_ce": 0.58242, "decode.acc_seg": 80.09418, "loss": 0.58242, "time": 0.1658}
+{"mode": "train", "epoch": 2, "iter": 1000, "lr": 0.00015, "memory": 19750, "data_time": 0.00717, "decode.loss_ce": 0.50039, "decode.acc_seg": 82.35546, "loss": 0.50039, "time": 0.16939}
+{"mode": "train", "epoch": 2, "iter": 1050, "lr": 0.00015, "memory": 19750, "data_time": 0.00674, "decode.loss_ce": 0.56778, "decode.acc_seg": 80.58647, "loss": 0.56778, "time": 0.16986}
+{"mode": "train", "epoch": 2, "iter": 1100, "lr": 0.00015, "memory": 19750, "data_time": 0.00664, "decode.loss_ce": 0.49441, "decode.acc_seg": 82.70061, "loss": 0.49441, "time": 0.16401}
+{"mode": "train", "epoch": 2, "iter": 1150, "lr": 0.00015, "memory": 19750, "data_time": 0.00655, "decode.loss_ce": 0.51378, "decode.acc_seg": 82.00246, "loss": 0.51378, "time": 0.16508}
+{"mode": "train", "epoch": 2, "iter": 1200, "lr": 0.00015, "memory": 19750, "data_time": 0.0066, "decode.loss_ce": 0.50145, "decode.acc_seg": 82.46899, "loss": 0.50145, "time": 0.17358}
+{"mode": "train", "epoch": 2, "iter": 1250, "lr": 0.00015, "memory": 19750, "data_time": 0.00695, "decode.loss_ce": 0.47563, "decode.acc_seg": 83.09324, "loss": 0.47563, "time": 0.16785}
+{"mode": "train", "epoch": 3, "iter": 1300, "lr": 0.00015, "memory": 19750, "data_time": 0.05795, "decode.loss_ce": 0.4762, "decode.acc_seg": 83.24759, "loss": 0.4762, "time": 0.23026}
+{"mode": "train", "epoch": 3, "iter": 1350, "lr": 0.00015, "memory": 19750, "data_time": 0.00627, "decode.loss_ce": 0.44346, "decode.acc_seg": 84.0603, "loss": 0.44346, "time": 0.17177}
+{"mode": "train", "epoch": 3, "iter": 1400, "lr": 0.00015, "memory": 19750, "data_time": 0.00717, "decode.loss_ce": 0.47084, "decode.acc_seg": 83.08637, "loss": 0.47084, "time": 0.16471}
+{"mode": "train", "epoch": 3, "iter": 1450, "lr": 0.00015, "memory": 19750, "data_time": 0.00639, "decode.loss_ce": 0.48639, "decode.acc_seg": 83.01241, "loss": 0.48639, "time": 0.16727}
+{"mode": "train", "epoch": 3, "iter": 1500, "lr": 0.00015, "memory": 19750, "data_time": 0.00674, "decode.loss_ce": 0.41735, "decode.acc_seg": 84.91397, "loss": 0.41735, "time": 0.16454}
+{"mode": "train", "epoch": 3, "iter": 1550, "lr": 0.00015, "memory": 19750, "data_time": 0.00674, "decode.loss_ce": 0.43834, "decode.acc_seg": 84.30104, "loss": 0.43834, "time": 0.16501}
+{"mode": "train", "epoch": 3, "iter": 1600, "lr": 0.00015, "memory": 19750, "data_time": 0.00687, "decode.loss_ce": 0.42182, "decode.acc_seg": 84.60363, "loss": 0.42182, "time": 0.16472}
+{"mode": "train", "epoch": 3, "iter": 1650, "lr": 0.00015, "memory": 19750, "data_time": 0.00682, "decode.loss_ce": 0.44227, "decode.acc_seg": 84.10865, "loss": 0.44227, "time": 0.17544}
+{"mode": "train", "epoch": 3, "iter": 1700, "lr": 0.00015, "memory": 19750, "data_time": 0.00676, "decode.loss_ce": 0.42305, "decode.acc_seg": 84.69608, "loss": 0.42305, "time": 0.16529}
+{"mode": "train", "epoch": 3, "iter": 1750, "lr": 0.00015, "memory": 19750, "data_time": 0.00635, "decode.loss_ce": 0.40065, "decode.acc_seg": 85.28233, "loss": 0.40065, "time": 0.17015}
+{"mode": "train", "epoch": 3, "iter": 1800, "lr": 0.00015, "memory": 19750, "data_time": 0.00687, "decode.loss_ce": 0.44377, "decode.acc_seg": 84.24874, "loss": 0.44377, "time": 0.17315}
+{"mode": "train", "epoch": 3, "iter": 1850, "lr": 0.00015, "memory": 19750, "data_time": 0.00669, "decode.loss_ce": 0.38964, "decode.acc_seg": 85.73132, "loss": 0.38964, "time": 0.16534}
+{"mode": "train", "epoch": 4, "iter": 1900, "lr": 0.00015, "memory": 19750, "data_time": 0.05387, "decode.loss_ce": 0.40727, "decode.acc_seg": 85.29138, "loss": 0.40727, "time": 0.21898}
+{"mode": "train", "epoch": 4, "iter": 1950, "lr": 0.00015, "memory": 19750, "data_time": 0.00683, "decode.loss_ce": 0.40309, "decode.acc_seg": 85.37785, "loss": 0.40309, "time": 0.16413}
+{"mode": "train", "epoch": 4, "iter": 2000, "lr": 0.00015, "memory": 19750, "data_time": 0.00651, "decode.loss_ce": 0.39313, "decode.acc_seg": 85.66311, "loss": 0.39313, "time": 0.16628}
+{"mode": "train", "epoch": 4, "iter": 2050, "lr": 0.00015, "memory": 19750, "data_time": 0.00657, "decode.loss_ce": 0.37103, "decode.acc_seg": 86.2932, "loss": 0.37103, "time": 0.1699}
+{"mode": "train", "epoch": 4, "iter": 2100, "lr": 0.00015, "memory": 19750, "data_time": 0.00694, "decode.loss_ce": 0.41258, "decode.acc_seg": 84.80687, "loss": 0.41258, "time": 0.16672}
+{"mode": "train", "epoch": 4, "iter": 2150, "lr": 0.00015, "memory": 19750, "data_time": 0.00683, "decode.loss_ce": 0.37237, "decode.acc_seg": 85.9416, "loss": 0.37237, "time": 0.16772}
+{"mode": "train", "epoch": 4, "iter": 2200, "lr": 0.00015, "memory": 19750, "data_time": 0.0067, "decode.loss_ce": 0.36418, "decode.acc_seg": 86.28289, "loss": 0.36418, "time": 0.16504}
+{"mode": "train", "epoch": 4, "iter": 2250, "lr": 0.00015, "memory": 19750, "data_time": 0.00682, "decode.loss_ce": 0.37703, "decode.acc_seg": 86.25602, "loss": 0.37703, "time": 0.16477}
+{"mode": "train", "epoch": 4, "iter": 2300, "lr": 0.00015, "memory": 19750, "data_time": 0.00658, "decode.loss_ce": 0.38606, "decode.acc_seg": 85.83164, "loss": 0.38606, "time": 0.17027}
+{"mode": "train", "epoch": 4, "iter": 2350, "lr": 0.00015, "memory": 19750, "data_time": 0.00701, "decode.loss_ce": 0.36377, "decode.acc_seg": 86.44438, "loss": 0.36377, "time": 0.16446}
+{"mode": "train", "epoch": 4, "iter": 2400, "lr": 0.00015, "memory": 19750, "data_time": 0.0069, "decode.loss_ce": 0.37793, "decode.acc_seg": 85.84639, "loss": 0.37793, "time": 0.16814}
+{"mode": "train", "epoch": 4, "iter": 2450, "lr": 0.00015, "memory": 19750, "data_time": 0.00671, "decode.loss_ce": 0.38427, "decode.acc_seg": 85.96406, "loss": 0.38427, "time": 0.16932}
+{"mode": "train", "epoch": 4, "iter": 2500, "lr": 0.00015, "memory": 19750, "data_time": 0.00823, "decode.loss_ce": 0.39177, "decode.acc_seg": 85.56283, "loss": 0.39177, "time": 0.17777}
+{"mode": "train", "epoch": 5, "iter": 2550, "lr": 0.00015, "memory": 19750, "data_time": 0.05398, "decode.loss_ce": 0.38108, "decode.acc_seg": 86.09614, "loss": 0.38108, "time": 0.21458}
+{"mode": "train", "epoch": 5, "iter": 2600, "lr": 0.00015, "memory": 19750, "data_time": 0.00655, "decode.loss_ce": 0.36287, "decode.acc_seg": 86.33698, "loss": 0.36287, "time": 0.16329}
+{"mode": "train", "epoch": 5, "iter": 2650, "lr": 0.00015, "memory": 19750, "data_time": 0.00676, "decode.loss_ce": 0.35487, "decode.acc_seg": 86.52721, "loss": 0.35487, "time": 0.16696}
+{"mode": "train", "epoch": 5, "iter": 2700, "lr": 0.00015, "memory": 19750, "data_time": 0.00703, "decode.loss_ce": 0.34173, "decode.acc_seg": 86.93155, "loss": 0.34173, "time": 0.16757}
+{"mode": "train", "epoch": 5, "iter": 2750, "lr": 0.00015, "memory": 19750, "data_time": 0.00765, "decode.loss_ce": 0.367, "decode.acc_seg": 86.25868, "loss": 0.367, "time": 0.17299}
+{"mode": "train", "epoch": 5, "iter": 2800, "lr": 0.00015, "memory": 19750, "data_time": 0.00677, "decode.loss_ce": 0.34542, "decode.acc_seg": 87.12672, "loss": 0.34542, "time": 0.16933}
+{"mode": "train", "epoch": 5, "iter": 2850, "lr": 0.00015, "memory": 19750, "data_time": 0.00673, "decode.loss_ce": 0.33885, "decode.acc_seg": 86.94783, "loss": 0.33885, "time": 0.17544}
+{"mode": "train", "epoch": 5, "iter": 2900, "lr": 0.00015, "memory": 19750, "data_time": 0.00719, "decode.loss_ce": 0.37236, "decode.acc_seg": 86.18583, "loss": 0.37236, "time": 0.16504}
+{"mode": "train", "epoch": 5, "iter": 2950, "lr": 0.00015, "memory": 19750, "data_time": 0.00744, "decode.loss_ce": 0.35252, "decode.acc_seg": 86.69874, "loss": 0.35252, "time": 0.16951}
+{"mode": "train", "epoch": 5, "iter": 3000, "lr": 0.00015, "memory": 19750, "data_time": 0.00674, "decode.loss_ce": 0.35913, "decode.acc_seg": 86.69249, "loss": 0.35913, "time": 0.16899}
+{"mode": "train", "epoch": 5, "iter": 3050, "lr": 0.00015, "memory": 19750, "data_time": 0.00722, "decode.loss_ce": 0.34791, "decode.acc_seg": 86.99021, "loss": 0.34791, "time": 0.17252}
+{"mode": "train", "epoch": 5, "iter": 3100, "lr": 0.00015, "memory": 19750, "data_time": 0.007, "decode.loss_ce": 0.3334, "decode.acc_seg": 87.31124, "loss": 0.3334, "time": 0.16188}
+{"mode": "train", "epoch": 5, "iter": 3150, "lr": 0.00015, "memory": 19750, "data_time": 0.00757, "decode.loss_ce": 0.34055, "decode.acc_seg": 87.3015, "loss": 0.34055, "time": 0.16971}
+{"mode": "train", "epoch": 6, "iter": 3200, "lr": 0.00015, "memory": 19750, "data_time": 0.05379, "decode.loss_ce": 0.36563, "decode.acc_seg": 86.3573, "loss": 0.36563, "time": 0.21416}
+{"mode": "train", "epoch": 6, "iter": 3250, "lr": 0.00015, "memory": 19750, "data_time": 0.00682, "decode.loss_ce": 0.31579, "decode.acc_seg": 88.11451, "loss": 0.31579, "time": 0.17126}
+{"mode": "train", "epoch": 6, "iter": 3300, "lr": 0.00015, "memory": 19750, "data_time": 0.00661, "decode.loss_ce": 0.36865, "decode.acc_seg": 86.31161, "loss": 0.36865, "time": 0.16823}
+{"mode": "train", "epoch": 6, "iter": 3350, "lr": 0.00015, "memory": 19750, "data_time": 0.00672, "decode.loss_ce": 0.34042, "decode.acc_seg": 87.06849, "loss": 0.34042, "time": 0.17189}
+{"mode": "train", "epoch": 6, "iter": 3400, "lr": 0.00015, "memory": 19750, "data_time": 0.00766, "decode.loss_ce": 0.33385, "decode.acc_seg": 87.29016, "loss": 0.33385, "time": 0.17741}
+{"mode": "train", "epoch": 6, "iter": 3450, "lr": 0.00015, "memory": 19750, "data_time": 0.00694, "decode.loss_ce": 0.33919, "decode.acc_seg": 86.97043, "loss": 0.33919, "time": 0.16819}
+{"mode": "train", "epoch": 6, "iter": 3500, "lr": 0.00015, "memory": 19750, "data_time": 0.00677, "decode.loss_ce": 0.33378, "decode.acc_seg": 87.10703, "loss": 0.33378, "time": 0.16973}
+{"mode": "train", "epoch": 6, "iter": 3550, "lr": 0.00015, "memory": 19750, "data_time": 0.00703, "decode.loss_ce": 0.32859, "decode.acc_seg": 87.48225, "loss": 0.32859, "time": 0.17087}
+{"mode": "train", "epoch": 6, "iter": 3600, "lr": 0.00015, "memory": 19750, "data_time": 0.00689, "decode.loss_ce": 0.34604, "decode.acc_seg": 86.52245, "loss": 0.34604, "time": 0.17393}
+{"mode": "train", "epoch": 6, "iter": 3650, "lr": 0.00015, "memory": 19750, "data_time": 0.00642, "decode.loss_ce": 0.34195, "decode.acc_seg": 87.02483, "loss": 0.34195, "time": 0.17442}
+{"mode": "train", "epoch": 6, "iter": 3700, "lr": 0.00015, "memory": 19750, "data_time": 0.00727, "decode.loss_ce": 0.3416, "decode.acc_seg": 87.16491, "loss": 0.3416, "time": 0.17331}
+{"mode": "train", "epoch": 6, "iter": 3750, "lr": 0.00015, "memory": 19750, "data_time": 0.00681, "decode.loss_ce": 0.34954, "decode.acc_seg": 86.81752, "loss": 0.34954, "time": 0.16843}
+{"mode": "train", "epoch": 7, "iter": 3800, "lr": 0.00015, "memory": 19750, "data_time": 0.05296, "decode.loss_ce": 0.34112, "decode.acc_seg": 87.0386, "loss": 0.34112, "time": 0.21411}
+{"mode": "train", "epoch": 7, "iter": 3850, "lr": 0.00015, "memory": 19750, "data_time": 0.00656, "decode.loss_ce": 0.34441, "decode.acc_seg": 86.98191, "loss": 0.34441, "time": 0.16859}
+{"mode": "train", "epoch": 7, "iter": 3900, "lr": 0.00015, "memory": 19750, "data_time": 0.00633, "decode.loss_ce": 0.3237, "decode.acc_seg": 87.60705, "loss": 0.3237, "time": 0.17053}
+{"mode": "train", "epoch": 7, "iter": 3950, "lr": 0.00015, "memory": 19750, "data_time": 0.00643, "decode.loss_ce": 0.34264, "decode.acc_seg": 87.06793, "loss": 0.34264, "time": 0.16719}
+{"mode": "train", "epoch": 7, "iter": 4000, "lr": 0.00015, "memory": 19750, "data_time": 0.00677, "decode.loss_ce": 0.33751, "decode.acc_seg": 86.86388, "loss": 0.33751, "time": 0.17013}
+{"mode": "train", "epoch": 7, "iter": 4050, "lr": 0.00015, "memory": 19750, "data_time": 0.00773, "decode.loss_ce": 0.32156, "decode.acc_seg": 87.75299, "loss": 0.32156, "time": 0.17056}
+{"mode": "train", "epoch": 7, "iter": 4100, "lr": 0.00015, "memory": 19750, "data_time": 0.00679, "decode.loss_ce": 0.33281, "decode.acc_seg": 87.0733, "loss": 0.33281, "time": 0.16674}
+{"mode": "train", "epoch": 7, "iter": 4150, "lr": 0.00015, "memory": 19750, "data_time": 0.00678, "decode.loss_ce": 0.29832, "decode.acc_seg": 88.33082, "loss": 0.29832, "time": 0.17049}
+{"mode": "train", "epoch": 7, "iter": 4200, "lr": 0.00015, "memory": 19750, "data_time": 0.00734, "decode.loss_ce": 0.31208, "decode.acc_seg": 88.04595, "loss": 0.31208, "time": 0.17608}
+{"mode": "train", "epoch": 7, "iter": 4250, "lr": 0.00015, "memory": 19750, "data_time": 0.00708, "decode.loss_ce": 0.33336, "decode.acc_seg": 87.25302, "loss": 0.33336, "time": 0.16545}
+{"mode": "train", "epoch": 7, "iter": 4300, "lr": 0.00015, "memory": 19750, "data_time": 0.00665, "decode.loss_ce": 0.32272, "decode.acc_seg": 87.632, "loss": 0.32272, "time": 0.16879}
+{"mode": "train", "epoch": 7, "iter": 4350, "lr": 0.00015, "memory": 19750, "data_time": 0.00716, "decode.loss_ce": 0.32937, "decode.acc_seg": 87.48962, "loss": 0.32937, "time": 0.17403}
+{"mode": "train", "epoch": 7, "iter": 4400, "lr": 0.00015, "memory": 19750, "data_time": 0.00672, "decode.loss_ce": 0.30072, "decode.acc_seg": 88.43203, "loss": 0.30072, "time": 0.17436}
+{"mode": "train", "epoch": 8, "iter": 4450, "lr": 0.00015, "memory": 19750, "data_time": 0.05521, "decode.loss_ce": 0.33604, "decode.acc_seg": 87.22331, "loss": 0.33604, "time": 0.21903}
+{"mode": "train", "epoch": 8, "iter": 4500, "lr": 0.00015, "memory": 19750, "data_time": 0.00658, "decode.loss_ce": 0.34547, "decode.acc_seg": 86.72177, "loss": 0.34547, "time": 0.1719}
+{"mode": "train", "epoch": 8, "iter": 4550, "lr": 0.00015, "memory": 19750, "data_time": 0.00647, "decode.loss_ce": 0.3248, "decode.acc_seg": 87.55375, "loss": 0.3248, "time": 0.16413}
+{"mode": "train", "epoch": 8, "iter": 4600, "lr": 0.00015, "memory": 19750, "data_time": 0.00625, "decode.loss_ce": 0.32307, "decode.acc_seg": 87.64411, "loss": 0.32307, "time": 0.16697}
+{"mode": "train", "epoch": 8, "iter": 4650, "lr": 0.00015, "memory": 19750, "data_time": 0.00598, "decode.loss_ce": 0.31148, "decode.acc_seg": 87.99871, "loss": 0.31148, "time": 0.18192}
+{"mode": "train", "epoch": 8, "iter": 4700, "lr": 0.00015, "memory": 19750, "data_time": 0.00623, "decode.loss_ce": 0.31103, "decode.acc_seg": 87.92522, "loss": 0.31103, "time": 0.17117}
+{"mode": "train", "epoch": 8, "iter": 4750, "lr": 0.00015, "memory": 19750, "data_time": 0.00687, "decode.loss_ce": 0.31458, "decode.acc_seg": 87.946, "loss": 0.31458, "time": 0.17047}
+{"mode": "train", "epoch": 8, "iter": 4800, "lr": 0.00015, "memory": 19750, "data_time": 0.00641, "decode.loss_ce": 0.32922, "decode.acc_seg": 87.34764, "loss": 0.32922, "time": 0.17031}
+{"mode": "train", "epoch": 8, "iter": 4850, "lr": 0.00015, "memory": 19750, "data_time": 0.00626, "decode.loss_ce": 0.3118, "decode.acc_seg": 87.8422, "loss": 0.3118, "time": 0.16599}
+{"mode": "train", "epoch": 8, "iter": 4900, "lr": 0.00015, "memory": 19750, "data_time": 0.00597, "decode.loss_ce": 0.29622, "decode.acc_seg": 88.33316, "loss": 0.29622, "time": 0.17134}
+{"mode": "train", "epoch": 8, "iter": 4950, "lr": 0.00015, "memory": 19750, "data_time": 0.00671, "decode.loss_ce": 0.31746, "decode.acc_seg": 87.63461, "loss": 0.31746, "time": 0.1665}
+{"mode": "train", "epoch": 8, "iter": 5000, "lr": 0.00015, "memory": 19750, "data_time": 0.00622, "decode.loss_ce": 0.30714, "decode.acc_seg": 88.24625, "loss": 0.30714, "time": 0.17579}
+{"mode": "train", "epoch": 9, "iter": 5050, "lr": 0.00015, "memory": 19750, "data_time": 0.05469, "decode.loss_ce": 0.29698, "decode.acc_seg": 88.53975, "loss": 0.29698, "time": 0.21976}
+{"mode": "train", "epoch": 9, "iter": 5100, "lr": 0.00015, "memory": 19750, "data_time": 0.00639, "decode.loss_ce": 0.30807, "decode.acc_seg": 88.01767, "loss": 0.30807, "time": 0.16527}
+{"mode": "train", "epoch": 9, "iter": 5150, "lr": 0.00015, "memory": 19750, "data_time": 0.00624, "decode.loss_ce": 0.3155, "decode.acc_seg": 88.08894, "loss": 0.3155, "time": 0.17069}
+{"mode": "train", "epoch": 9, "iter": 5200, "lr": 0.00015, "memory": 19750, "data_time": 0.00592, "decode.loss_ce": 0.29819, "decode.acc_seg": 88.41932, "loss": 0.29819, "time": 0.16355}
+{"mode": "train", "epoch": 9, "iter": 5250, "lr": 0.00015, "memory": 19750, "data_time": 0.00656, "decode.loss_ce": 0.2976, "decode.acc_seg": 88.17167, "loss": 0.2976, "time": 0.17297}
+{"mode": "train", "epoch": 9, "iter": 5300, "lr": 0.00015, "memory": 19750, "data_time": 0.00595, "decode.loss_ce": 0.31431, "decode.acc_seg": 87.96738, "loss": 0.31431, "time": 0.17553}
+{"mode": "train", "epoch": 9, "iter": 5350, "lr": 0.00015, "memory": 19750, "data_time": 0.00632, "decode.loss_ce": 0.31812, "decode.acc_seg": 87.89434, "loss": 0.31812, "time": 0.16414}
+{"mode": "train", "epoch": 9, "iter": 5400, "lr": 0.00015, "memory": 19750, "data_time": 0.00631, "decode.loss_ce": 0.32172, "decode.acc_seg": 87.40795, "loss": 0.32172, "time": 0.16415}
+{"mode": "train", "epoch": 9, "iter": 5450, "lr": 0.00015, "memory": 19750, "data_time": 0.00636, "decode.loss_ce": 0.29955, "decode.acc_seg": 88.12618, "loss": 0.29955, "time": 0.1643}
+{"mode": "train", "epoch": 9, "iter": 5500, "lr": 0.00015, "memory": 19750, "data_time": 0.00622, "decode.loss_ce": 0.29215, "decode.acc_seg": 88.49556, "loss": 0.29215, "time": 0.16871}
+{"mode": "train", "epoch": 9, "iter": 5550, "lr": 0.00015, "memory": 19750, "data_time": 0.00596, "decode.loss_ce": 0.32296, "decode.acc_seg": 87.6716, "loss": 0.32296, "time": 0.17221}
+{"mode": "train", "epoch": 9, "iter": 5600, "lr": 0.00015, "memory": 19750, "data_time": 0.00725, "decode.loss_ce": 0.30732, "decode.acc_seg": 88.08342, "loss": 0.30732, "time": 0.17205}
+{"mode": "train", "epoch": 9, "iter": 5650, "lr": 0.00015, "memory": 19750, "data_time": 0.00639, "decode.loss_ce": 0.31748, "decode.acc_seg": 87.74626, "loss": 0.31748, "time": 0.17152}
+{"mode": "train", "epoch": 10, "iter": 5700, "lr": 0.00015, "memory": 19750, "data_time": 0.05224, "decode.loss_ce": 0.2968, "decode.acc_seg": 88.33484, "loss": 0.2968, "time": 0.22297}
+{"mode": "train", "epoch": 10, "iter": 5750, "lr": 0.00015, "memory": 19750, "data_time": 0.00603, "decode.loss_ce": 0.2844, "decode.acc_seg": 88.79802, "loss": 0.2844, "time": 0.16865}
+{"mode": "train", "epoch": 10, "iter": 5800, "lr": 0.00015, "memory": 19750, "data_time": 0.00642, "decode.loss_ce": 0.30396, "decode.acc_seg": 88.12515, "loss": 0.30396, "time": 0.16799}
+{"mode": "train", "epoch": 10, "iter": 5850, "lr": 0.00015, "memory": 19750, "data_time": 0.00617, "decode.loss_ce": 0.30582, "decode.acc_seg": 87.97682, "loss": 0.30582, "time": 0.1653}
+{"mode": "train", "epoch": 10, "iter": 5900, "lr": 0.00015, "memory": 19750, "data_time": 0.00661, "decode.loss_ce": 0.29512, "decode.acc_seg": 88.46608, "loss": 0.29512, "time": 0.16527}
+{"mode": "train", "epoch": 10, "iter": 5950, "lr": 0.00015, "memory": 19750, "data_time": 0.00597, "decode.loss_ce": 0.31263, "decode.acc_seg": 87.92211, "loss": 0.31263, "time": 0.17091}
+{"mode": "train", "epoch": 10, "iter": 6000, "lr": 0.00015, "memory": 19750, "data_time": 0.00653, "decode.loss_ce": 0.29545, "decode.acc_seg": 88.35079, "loss": 0.29545, "time": 0.16982}
+{"mode": "train", "epoch": 10, "iter": 6050, "lr": 0.00015, "memory": 19750, "data_time": 0.00631, "decode.loss_ce": 0.28433, "decode.acc_seg": 88.93129, "loss": 0.28433, "time": 0.16613}
+{"mode": "train", "epoch": 10, "iter": 6100, "lr": 0.00015, "memory": 19750, "data_time": 0.00695, "decode.loss_ce": 0.30451, "decode.acc_seg": 88.19103, "loss": 0.30451, "time": 0.16886}
+{"mode": "train", "epoch": 10, "iter": 6150, "lr": 0.00015, "memory": 19750, "data_time": 0.00623, "decode.loss_ce": 0.30189, "decode.acc_seg": 88.29146, "loss": 0.30189, "time": 0.16576}
+{"mode": "train", "epoch": 10, "iter": 6200, "lr": 0.00015, "memory": 19750, "data_time": 0.0065, "decode.loss_ce": 0.30363, "decode.acc_seg": 88.20825, "loss": 0.30363, "time": 0.17004}
+{"mode": "train", "epoch": 10, "iter": 6250, "lr": 0.00015, "memory": 19750, "data_time": 0.00627, "decode.loss_ce": 0.30805, "decode.acc_seg": 88.04902, "loss": 0.30805, "time": 0.16293}
+{"mode": "train", "epoch": 10, "iter": 6300, "lr": 0.00015, "memory": 19750, "data_time": 0.00655, "decode.loss_ce": 0.30825, "decode.acc_seg": 88.29885, "loss": 0.30825, "time": 0.16624}
+{"mode": "train", "epoch": 11, "iter": 6350, "lr": 0.00015, "memory": 19750, "data_time": 0.0546, "decode.loss_ce": 0.30881, "decode.acc_seg": 87.92228, "loss": 0.30881, "time": 0.21774}
+{"mode": "train", "epoch": 11, "iter": 6400, "lr": 0.00015, "memory": 19750, "data_time": 0.0062, "decode.loss_ce": 0.30339, "decode.acc_seg": 88.08356, "loss": 0.30339, "time": 0.17795}
+{"mode": "train", "epoch": 11, "iter": 6450, "lr": 0.00015, "memory": 19750, "data_time": 0.00668, "decode.loss_ce": 0.28164, "decode.acc_seg": 88.88014, "loss": 0.28164, "time": 0.16356}
+{"mode": "train", "epoch": 11, "iter": 6500, "lr": 0.00015, "memory": 19750, "data_time": 0.00635, "decode.loss_ce": 0.29401, "decode.acc_seg": 88.49332, "loss": 0.29401, "time": 0.17147}
+{"mode": "train", "epoch": 11, "iter": 6550, "lr": 0.00015, "memory": 19750, "data_time": 0.00621, "decode.loss_ce": 0.29546, "decode.acc_seg": 88.55466, "loss": 0.29546, "time": 0.17142}
+{"mode": "train", "epoch": 11, "iter": 6600, "lr": 0.00015, "memory": 19750, "data_time": 0.00676, "decode.loss_ce": 0.32099, "decode.acc_seg": 87.74059, "loss": 0.32099, "time": 0.16338}
+{"mode": "train", "epoch": 11, "iter": 6650, "lr": 0.00015, "memory": 19750, "data_time": 0.00609, "decode.loss_ce": 0.31259, "decode.acc_seg": 87.85196, "loss": 0.31259, "time": 0.17431}
+{"mode": "train", "epoch": 11, "iter": 6700, "lr": 0.00015, "memory": 19750, "data_time": 0.00688, "decode.loss_ce": 0.28936, "decode.acc_seg": 88.58093, "loss": 0.28936, "time": 0.16549}
+{"mode": "train", "epoch": 11, "iter": 6750, "lr": 0.00015, "memory": 19750, "data_time": 0.00625, "decode.loss_ce": 0.31332, "decode.acc_seg": 88.07452, "loss": 0.31332, "time": 0.17283}
+{"mode": "train", "epoch": 11, "iter": 6800, "lr": 0.00015, "memory": 19750, "data_time": 0.00632, "decode.loss_ce": 0.30062, "decode.acc_seg": 88.2473, "loss": 0.30062, "time": 0.16795}
+{"mode": "train", "epoch": 11, "iter": 6850, "lr": 0.00015, "memory": 19750, "data_time": 0.00607, "decode.loss_ce": 0.30173, "decode.acc_seg": 88.37736, "loss": 0.30173, "time": 0.17001}
+{"mode": "train", "epoch": 11, "iter": 6900, "lr": 0.00015, "memory": 19750, "data_time": 0.00638, "decode.loss_ce": 0.28008, "decode.acc_seg": 88.99362, "loss": 0.28008, "time": 0.16775}
+{"mode": "train", "epoch": 12, "iter": 6950, "lr": 0.00015, "memory": 19750, "data_time": 0.05632, "decode.loss_ce": 0.28848, "decode.acc_seg": 88.79085, "loss": 0.28848, "time": 0.22683}
+{"mode": "train", "epoch": 12, "iter": 7000, "lr": 0.00015, "memory": 19750, "data_time": 0.00659, "decode.loss_ce": 0.29346, "decode.acc_seg": 88.4865, "loss": 0.29346, "time": 0.16459}
+{"mode": "train", "epoch": 12, "iter": 7050, "lr": 0.00015, "memory": 19750, "data_time": 0.00658, "decode.loss_ce": 0.28864, "decode.acc_seg": 88.43235, "loss": 0.28864, "time": 0.17538}
+{"mode": "train", "epoch": 12, "iter": 7100, "lr": 0.00015, "memory": 19750, "data_time": 0.00727, "decode.loss_ce": 0.29303, "decode.acc_seg": 88.31878, "loss": 0.29303, "time": 0.16954}
+{"mode": "train", "epoch": 12, "iter": 7150, "lr": 0.00015, "memory": 19750, "data_time": 0.00733, "decode.loss_ce": 0.2804, "decode.acc_seg": 89.1287, "loss": 0.2804, "time": 0.16794}
+{"mode": "train", "epoch": 12, "iter": 7200, "lr": 0.00015, "memory": 19750, "data_time": 0.00645, "decode.loss_ce": 0.29176, "decode.acc_seg": 88.75452, "loss": 0.29176, "time": 0.16516}
+{"mode": "train", "epoch": 12, "iter": 7250, "lr": 0.00015, "memory": 19750, "data_time": 0.00696, "decode.loss_ce": 0.27544, "decode.acc_seg": 89.21126, "loss": 0.27544, "time": 0.16841}
+{"mode": "train", "epoch": 12, "iter": 7300, "lr": 0.00015, "memory": 19750, "data_time": 0.00684, "decode.loss_ce": 0.28735, "decode.acc_seg": 88.76757, "loss": 0.28735, "time": 0.16692}
+{"mode": "train", "epoch": 12, "iter": 7350, "lr": 0.00015, "memory": 19750, "data_time": 0.00643, "decode.loss_ce": 0.3013, "decode.acc_seg": 88.42977, "loss": 0.3013, "time": 0.1705}
+{"mode": "train", "epoch": 12, "iter": 7400, "lr": 0.00015, "memory": 19750, "data_time": 0.00599, "decode.loss_ce": 0.27759, "decode.acc_seg": 88.99402, "loss": 0.27759, "time": 0.17244}
+{"mode": "train", "epoch": 12, "iter": 7450, "lr": 0.00015, "memory": 19750, "data_time": 0.00631, "decode.loss_ce": 0.29072, "decode.acc_seg": 88.72828, "loss": 0.29072, "time": 0.16272}
+{"mode": "train", "epoch": 12, "iter": 7500, "lr": 0.00015, "memory": 19750, "data_time": 0.00608, "decode.loss_ce": 0.28911, "decode.acc_seg": 88.82428, "loss": 0.28911, "time": 0.16917}
+{"mode": "train", "epoch": 12, "iter": 7550, "lr": 0.00015, "memory": 19750, "data_time": 0.00694, "decode.loss_ce": 0.26827, "decode.acc_seg": 89.13525, "loss": 0.26827, "time": 0.1668}
+{"mode": "train", "epoch": 13, "iter": 7600, "lr": 0.00015, "memory": 19750, "data_time": 0.05774, "decode.loss_ce": 0.2865, "decode.acc_seg": 88.82241, "loss": 0.2865, "time": 0.22449}
+{"mode": "train", "epoch": 13, "iter": 7650, "lr": 0.00015, "memory": 19750, "data_time": 0.00617, "decode.loss_ce": 0.29029, "decode.acc_seg": 88.68461, "loss": 0.29029, "time": 0.16747}
+{"mode": "train", "epoch": 13, "iter": 7700, "lr": 0.00015, "memory": 19750, "data_time": 0.00615, "decode.loss_ce": 0.27594, "decode.acc_seg": 88.94087, "loss": 0.27594, "time": 0.17627}
+{"mode": "train", "epoch": 13, "iter": 7750, "lr": 0.00015, "memory": 19750, "data_time": 0.00641, "decode.loss_ce": 0.29474, "decode.acc_seg": 88.3407, "loss": 0.29474, "time": 0.16154}
+{"mode": "train", "epoch": 13, "iter": 7800, "lr": 0.00015, "memory": 19750, "data_time": 0.0066, "decode.loss_ce": 0.29451, "decode.acc_seg": 88.5218, "loss": 0.29451, "time": 0.16433}
+{"mode": "train", "epoch": 13, "iter": 7850, "lr": 0.00015, "memory": 19750, "data_time": 0.00625, "decode.loss_ce": 0.29442, "decode.acc_seg": 88.53064, "loss": 0.29442, "time": 0.17425}
+{"mode": "train", "epoch": 13, "iter": 7900, "lr": 0.00015, "memory": 19750, "data_time": 0.0062, "decode.loss_ce": 0.27688, "decode.acc_seg": 89.11096, "loss": 0.27688, "time": 0.17571}
+{"mode": "train", "epoch": 13, "iter": 7950, "lr": 0.00015, "memory": 19750, "data_time": 0.0068, "decode.loss_ce": 0.28287, "decode.acc_seg": 88.81269, "loss": 0.28287, "time": 0.16731}
+{"mode": "train", "epoch": 13, "iter": 8000, "lr": 0.00015, "memory": 19750, "data_time": 0.00737, "decode.loss_ce": 0.28287, "decode.acc_seg": 88.78357, "loss": 0.28287, "time": 0.17956}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_184631.log

@@ -0,0 +1,1139 @@
+2023-03-04 18:46:31,289 - mmseg - INFO - Multi-processing start method is `None`
+2023-03-04 18:46:31,301 - mmseg - INFO - OpenCV num_threads is `128
+2023-03-04 18:46:31,301 - mmseg - INFO - OMP num threads is 1
+2023-03-04 18:46:31,368 - mmseg - INFO - Environment info:
+------------------------------------------------------------
+sys.platform: linux
+Python: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]
+CUDA available: True
+GPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB
+CUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch
+NVCC: Cuda compilation tools, release 11.6, V11.6.124
+GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)
+PyTorch: 1.13.1
+PyTorch compiling details: PyTorch built with:
+  - GCC 9.3
+  - C++ Version: 201402
+  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
+  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
+  - OpenMP 201511 (a.k.a. OpenMP 4.5)
+  - LAPACK is enabled (usually provided by MKL)
+  - NNPACK is enabled
+  - CPU capability usage: AVX2
+  - CUDA Runtime 11.6
+  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
+  - CuDNN 8.3.2  (built against CUDA 11.5)
+  - Magma 2.6.1
+  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 
+
+TorchVision: 0.14.1
+OpenCV: 4.7.0
+MMCV: 1.7.1
+MMCV Compiler: GCC 9.3
+MMCV CUDA Compiler: 11.6
+MMSegmentation: 0.30.0+6749699
+------------------------------------------------------------
+
+2023-03-04 18:46:31,368 - mmseg - INFO - Distributed training: True
+2023-03-04 18:46:32,081 - mmseg - INFO - Config:
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepLogits',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=166,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'
+gpu_ids = range(0, 8)
+auto_resume = True
+
+2023-03-04 18:46:38,332 - mmseg - INFO - Set random seed to 1082958590, deterministic: False
+2023-03-04 18:46:38,583 - mmseg - INFO - Parameters in backbone freezed!
+2023-03-04 18:46:38,583 - mmseg - INFO - Trainable parameters in SegformerHeadUnetFCHeadSingleStep: ['unet.init_conv.weight', 'unet.init_conv.bias', 'unet.time_mlp.1.weight', 'unet.time_mlp.1.bias', 'unet.time_mlp.3.weight', 'unet.time_mlp.3.bias', 'unet.downs.0.0.mlp.1.weight', 'unet.downs.0.0.mlp.1.bias', 'unet.downs.0.0.block1.proj.weight', 'unet.downs.0.0.block1.proj.bias', 'unet.downs.0.0.block1.norm.weight', 'unet.downs.0.0.block1.norm.bias', 'unet.downs.0.0.block2.proj.weight', 'unet.downs.0.0.block2.proj.bias', 'unet.downs.0.0.block2.norm.weight', 'unet.downs.0.0.block2.norm.bias', 'unet.downs.0.1.mlp.1.weight', 'unet.downs.0.1.mlp.1.bias', 'unet.downs.0.1.block1.proj.weight', 'unet.downs.0.1.block1.proj.bias', 'unet.downs.0.1.block1.norm.weight', 'unet.downs.0.1.block1.norm.bias', 'unet.downs.0.1.block2.proj.weight', 'unet.downs.0.1.block2.proj.bias', 'unet.downs.0.1.block2.norm.weight', 'unet.downs.0.1.block2.norm.bias', 'unet.downs.0.2.fn.fn.to_qkv.weight', 'unet.downs.0.2.fn.fn.to_out.0.weight', 'unet.downs.0.2.fn.fn.to_out.0.bias', 'unet.downs.0.2.fn.fn.to_out.1.g', 'unet.downs.0.2.fn.norm.g', 'unet.downs.0.3.weight', 'unet.downs.0.3.bias', 'unet.downs.1.0.mlp.1.weight', 'unet.downs.1.0.mlp.1.bias', 'unet.downs.1.0.block1.proj.weight', 'unet.downs.1.0.block1.proj.bias', 'unet.downs.1.0.block1.norm.weight', 'unet.downs.1.0.block1.norm.bias', 'unet.downs.1.0.block2.proj.weight', 'unet.downs.1.0.block2.proj.bias', 'unet.downs.1.0.block2.norm.weight', 'unet.downs.1.0.block2.norm.bias', 'unet.downs.1.1.mlp.1.weight', 'unet.downs.1.1.mlp.1.bias', 'unet.downs.1.1.block1.proj.weight', 'unet.downs.1.1.block1.proj.bias', 'unet.downs.1.1.block1.norm.weight', 'unet.downs.1.1.block1.norm.bias', 'unet.downs.1.1.block2.proj.weight', 'unet.downs.1.1.block2.proj.bias', 'unet.downs.1.1.block2.norm.weight', 'unet.downs.1.1.block2.norm.bias', 'unet.downs.1.2.fn.fn.to_qkv.weight', 'unet.downs.1.2.fn.fn.to_out.0.weight', 'unet.downs.1.2.fn.fn.to_out.0.bias', 'unet.downs.1.2.fn.fn.to_out.1.g', 'unet.downs.1.2.fn.norm.g', 'unet.downs.1.3.weight', 'unet.downs.1.3.bias', 'unet.downs.2.0.mlp.1.weight', 'unet.downs.2.0.mlp.1.bias', 'unet.downs.2.0.block1.proj.weight', 'unet.downs.2.0.block1.proj.bias', 'unet.downs.2.0.block1.norm.weight', 'unet.downs.2.0.block1.norm.bias', 'unet.downs.2.0.block2.proj.weight', 'unet.downs.2.0.block2.proj.bias', 'unet.downs.2.0.block2.norm.weight', 'unet.downs.2.0.block2.norm.bias', 'unet.downs.2.1.mlp.1.weight', 'unet.downs.2.1.mlp.1.bias', 'unet.downs.2.1.block1.proj.weight', 'unet.downs.2.1.block1.proj.bias', 'unet.downs.2.1.block1.norm.weight', 'unet.downs.2.1.block1.norm.bias', 'unet.downs.2.1.block2.proj.weight', 'unet.downs.2.1.block2.proj.bias', 'unet.downs.2.1.block2.norm.weight', 'unet.downs.2.1.block2.norm.bias', 'unet.downs.2.2.fn.fn.to_qkv.weight', 'unet.downs.2.2.fn.fn.to_out.0.weight', 'unet.downs.2.2.fn.fn.to_out.0.bias', 'unet.downs.2.2.fn.fn.to_out.1.g', 'unet.downs.2.2.fn.norm.g', 'unet.downs.2.3.weight', 'unet.downs.2.3.bias', 'unet.ups.0.0.mlp.1.weight', 'unet.ups.0.0.mlp.1.bias', 'unet.ups.0.0.block1.proj.weight', 'unet.ups.0.0.block1.proj.bias', 'unet.ups.0.0.block1.norm.weight', 'unet.ups.0.0.block1.norm.bias', 'unet.ups.0.0.block2.proj.weight', 'unet.ups.0.0.block2.proj.bias', 'unet.ups.0.0.block2.norm.weight', 'unet.ups.0.0.block2.norm.bias', 'unet.ups.0.0.res_conv.weight', 'unet.ups.0.0.res_conv.bias', 'unet.ups.0.1.mlp.1.weight', 'unet.ups.0.1.mlp.1.bias', 'unet.ups.0.1.block1.proj.weight', 'unet.ups.0.1.block1.proj.bias', 'unet.ups.0.1.block1.norm.weight', 'unet.ups.0.1.block1.norm.bias', 'unet.ups.0.1.block2.proj.weight', 'unet.ups.0.1.block2.proj.bias', 'unet.ups.0.1.block2.norm.weight', 'unet.ups.0.1.block2.norm.bias', 'unet.ups.0.1.res_conv.weight', 'unet.ups.0.1.res_conv.bias', 'unet.ups.0.2.fn.fn.to_qkv.weight', 'unet.ups.0.2.fn.fn.to_out.0.weight', 'unet.ups.0.2.fn.fn.to_out.0.bias', 'unet.ups.0.2.fn.fn.to_out.1.g', 'unet.ups.0.2.fn.norm.g', 'unet.ups.0.3.1.weight', 'unet.ups.0.3.1.bias', 'unet.ups.1.0.mlp.1.weight', 'unet.ups.1.0.mlp.1.bias', 'unet.ups.1.0.block1.proj.weight', 'unet.ups.1.0.block1.proj.bias', 'unet.ups.1.0.block1.norm.weight', 'unet.ups.1.0.block1.norm.bias', 'unet.ups.1.0.block2.proj.weight', 'unet.ups.1.0.block2.proj.bias', 'unet.ups.1.0.block2.norm.weight', 'unet.ups.1.0.block2.norm.bias', 'unet.ups.1.0.res_conv.weight', 'unet.ups.1.0.res_conv.bias', 'unet.ups.1.1.mlp.1.weight', 'unet.ups.1.1.mlp.1.bias', 'unet.ups.1.1.block1.proj.weight', 'unet.ups.1.1.block1.proj.bias', 'unet.ups.1.1.block1.norm.weight', 'unet.ups.1.1.block1.norm.bias', 'unet.ups.1.1.block2.proj.weight', 'unet.ups.1.1.block2.proj.bias', 'unet.ups.1.1.block2.norm.weight', 'unet.ups.1.1.block2.norm.bias', 'unet.ups.1.1.res_conv.weight', 'unet.ups.1.1.res_conv.bias', 'unet.ups.1.2.fn.fn.to_qkv.weight', 'unet.ups.1.2.fn.fn.to_out.0.weight', 'unet.ups.1.2.fn.fn.to_out.0.bias', 'unet.ups.1.2.fn.fn.to_out.1.g', 'unet.ups.1.2.fn.norm.g', 'unet.ups.1.3.1.weight', 'unet.ups.1.3.1.bias', 'unet.ups.2.0.mlp.1.weight', 'unet.ups.2.0.mlp.1.bias', 'unet.ups.2.0.block1.proj.weight', 'unet.ups.2.0.block1.proj.bias', 'unet.ups.2.0.block1.norm.weight', 'unet.ups.2.0.block1.norm.bias', 'unet.ups.2.0.block2.proj.weight', 'unet.ups.2.0.block2.proj.bias', 'unet.ups.2.0.block2.norm.weight', 'unet.ups.2.0.block2.norm.bias', 'unet.ups.2.0.res_conv.weight', 'unet.ups.2.0.res_conv.bias', 'unet.ups.2.1.mlp.1.weight', 'unet.ups.2.1.mlp.1.bias', 'unet.ups.2.1.block1.proj.weight', 'unet.ups.2.1.block1.proj.bias', 'unet.ups.2.1.block1.norm.weight', 'unet.ups.2.1.block1.norm.bias', 'unet.ups.2.1.block2.proj.weight', 'unet.ups.2.1.block2.proj.bias', 'unet.ups.2.1.block2.norm.weight', 'unet.ups.2.1.block2.norm.bias', 'unet.ups.2.1.res_conv.weight', 'unet.ups.2.1.res_conv.bias', 'unet.ups.2.2.fn.fn.to_qkv.weight', 'unet.ups.2.2.fn.fn.to_out.0.weight', 'unet.ups.2.2.fn.fn.to_out.0.bias', 'unet.ups.2.2.fn.fn.to_out.1.g', 'unet.ups.2.2.fn.norm.g', 'unet.ups.2.3.weight', 'unet.ups.2.3.bias', 'unet.mid_block1.mlp.1.weight', 'unet.mid_block1.mlp.1.bias', 'unet.mid_block1.block1.proj.weight', 'unet.mid_block1.block1.proj.bias', 'unet.mid_block1.block1.norm.weight', 'unet.mid_block1.block1.norm.bias', 'unet.mid_block1.block2.proj.weight', 'unet.mid_block1.block2.proj.bias', 'unet.mid_block1.block2.norm.weight', 'unet.mid_block1.block2.norm.bias', 'unet.mid_attn.fn.fn.to_qkv.weight', 'unet.mid_attn.fn.fn.to_out.weight', 'unet.mid_attn.fn.fn.to_out.bias', 'unet.mid_attn.fn.norm.g', 'unet.mid_block2.mlp.1.weight', 'unet.mid_block2.mlp.1.bias', 'unet.mid_block2.block1.proj.weight', 'unet.mid_block2.block1.proj.bias', 'unet.mid_block2.block1.norm.weight', 'unet.mid_block2.block1.norm.bias', 'unet.mid_block2.block2.proj.weight', 'unet.mid_block2.block2.proj.bias', 'unet.mid_block2.block2.norm.weight', 'unet.mid_block2.block2.norm.bias', 'unet.final_res_block.mlp.1.weight', 'unet.final_res_block.mlp.1.bias', 'unet.final_res_block.block1.proj.weight', 'unet.final_res_block.block1.proj.bias', 'unet.final_res_block.block1.norm.weight', 'unet.final_res_block.block1.norm.bias', 'unet.final_res_block.block2.proj.weight', 'unet.final_res_block.block2.proj.bias', 'unet.final_res_block.block2.norm.weight', 'unet.final_res_block.block2.norm.bias', 'unet.final_res_block.res_conv.weight', 'unet.final_res_block.res_conv.bias', 'unet.final_conv.weight', 'unet.final_conv.bias', 'conv_seg_new.weight', 'conv_seg_new.bias']
+2023-03-04 18:46:38,584 - mmseg - INFO - Parameters in decode_head freezed!
+2023-03-04 18:46:38,606 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 18:46:38,852 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: decode_head.conv_seg.weight, decode_head.conv_seg.bias, decode_head.convs.0.conv.weight, decode_head.convs.0.bn.weight, decode_head.convs.0.bn.bias, decode_head.convs.0.bn.running_mean, decode_head.convs.0.bn.running_var, decode_head.convs.0.bn.num_batches_tracked, decode_head.convs.1.conv.weight, decode_head.convs.1.bn.weight, decode_head.convs.1.bn.bias, decode_head.convs.1.bn.running_mean, decode_head.convs.1.bn.running_var, decode_head.convs.1.bn.num_batches_tracked, decode_head.convs.2.conv.weight, decode_head.convs.2.bn.weight, decode_head.convs.2.bn.bias, decode_head.convs.2.bn.running_mean, decode_head.convs.2.bn.running_var, decode_head.convs.2.bn.num_batches_tracked, decode_head.convs.3.conv.weight, decode_head.convs.3.bn.weight, decode_head.convs.3.bn.bias, decode_head.convs.3.bn.running_mean, decode_head.convs.3.bn.running_var, decode_head.convs.3.bn.num_batches_tracked, decode_head.fusion_conv.conv.weight, decode_head.fusion_conv.bn.weight, decode_head.fusion_conv.bn.bias, decode_head.fusion_conv.bn.running_mean, decode_head.fusion_conv.bn.running_var, decode_head.fusion_conv.bn.num_batches_tracked
+
+2023-03-04 18:46:38,865 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 18:46:39,075 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: backbone.layers.0.0.projection.weight, backbone.layers.0.0.projection.bias, backbone.layers.0.0.norm.weight, backbone.layers.0.0.norm.bias, backbone.layers.0.1.0.norm1.weight, backbone.layers.0.1.0.norm1.bias, backbone.layers.0.1.0.attn.attn.in_proj_weight, backbone.layers.0.1.0.attn.attn.in_proj_bias, backbone.layers.0.1.0.attn.attn.out_proj.weight, backbone.layers.0.1.0.attn.attn.out_proj.bias, backbone.layers.0.1.0.attn.sr.weight, backbone.layers.0.1.0.attn.sr.bias, backbone.layers.0.1.0.attn.norm.weight, backbone.layers.0.1.0.attn.norm.bias, backbone.layers.0.1.0.norm2.weight, backbone.layers.0.1.0.norm2.bias, backbone.layers.0.1.0.ffn.layers.0.weight, backbone.layers.0.1.0.ffn.layers.0.bias, backbone.layers.0.1.0.ffn.layers.1.weight, backbone.layers.0.1.0.ffn.layers.1.bias, backbone.layers.0.1.0.ffn.layers.4.weight, backbone.layers.0.1.0.ffn.layers.4.bias, backbone.layers.0.1.1.norm1.weight, backbone.layers.0.1.1.norm1.bias, backbone.layers.0.1.1.attn.attn.in_proj_weight, backbone.layers.0.1.1.attn.attn.in_proj_bias, backbone.layers.0.1.1.attn.attn.out_proj.weight, backbone.layers.0.1.1.attn.attn.out_proj.bias, backbone.layers.0.1.1.attn.sr.weight, backbone.layers.0.1.1.attn.sr.bias, backbone.layers.0.1.1.attn.norm.weight, backbone.layers.0.1.1.attn.norm.bias, backbone.layers.0.1.1.norm2.weight, backbone.layers.0.1.1.norm2.bias, backbone.layers.0.1.1.ffn.layers.0.weight, backbone.layers.0.1.1.ffn.layers.0.bias, backbone.layers.0.1.1.ffn.layers.1.weight, backbone.layers.0.1.1.ffn.layers.1.bias, backbone.layers.0.1.1.ffn.layers.4.weight, backbone.layers.0.1.1.ffn.layers.4.bias, backbone.layers.0.1.2.norm1.weight, backbone.layers.0.1.2.norm1.bias, backbone.layers.0.1.2.attn.attn.in_proj_weight, backbone.layers.0.1.2.attn.attn.in_proj_bias, backbone.layers.0.1.2.attn.attn.out_proj.weight, backbone.layers.0.1.2.attn.attn.out_proj.bias, backbone.layers.0.1.2.attn.sr.weight, backbone.layers.0.1.2.attn.sr.bias, backbone.layers.0.1.2.attn.norm.weight, backbone.layers.0.1.2.attn.norm.bias, backbone.layers.0.1.2.norm2.weight, backbone.layers.0.1.2.norm2.bias, backbone.layers.0.1.2.ffn.layers.0.weight, backbone.layers.0.1.2.ffn.layers.0.bias, backbone.layers.0.1.2.ffn.layers.1.weight, backbone.layers.0.1.2.ffn.layers.1.bias, backbone.layers.0.1.2.ffn.layers.4.weight, backbone.layers.0.1.2.ffn.layers.4.bias, backbone.layers.0.2.weight, backbone.layers.0.2.bias, backbone.layers.1.0.projection.weight, backbone.layers.1.0.projection.bias, backbone.layers.1.0.norm.weight, backbone.layers.1.0.norm.bias, backbone.layers.1.1.0.norm1.weight, backbone.layers.1.1.0.norm1.bias, backbone.layers.1.1.0.attn.attn.in_proj_weight, backbone.layers.1.1.0.attn.attn.in_proj_bias, backbone.layers.1.1.0.attn.attn.out_proj.weight, backbone.layers.1.1.0.attn.attn.out_proj.bias, backbone.layers.1.1.0.attn.sr.weight, backbone.layers.1.1.0.attn.sr.bias, backbone.layers.1.1.0.attn.norm.weight, backbone.layers.1.1.0.attn.norm.bias, backbone.layers.1.1.0.norm2.weight, backbone.layers.1.1.0.norm2.bias, backbone.layers.1.1.0.ffn.layers.0.weight, backbone.layers.1.1.0.ffn.layers.0.bias, backbone.layers.1.1.0.ffn.layers.1.weight, backbone.layers.1.1.0.ffn.layers.1.bias, backbone.layers.1.1.0.ffn.layers.4.weight, backbone.layers.1.1.0.ffn.layers.4.bias, backbone.layers.1.1.1.norm1.weight, backbone.layers.1.1.1.norm1.bias, backbone.layers.1.1.1.attn.attn.in_proj_weight, backbone.layers.1.1.1.attn.attn.in_proj_bias, backbone.layers.1.1.1.attn.attn.out_proj.weight, backbone.layers.1.1.1.attn.attn.out_proj.bias, backbone.layers.1.1.1.attn.sr.weight, backbone.layers.1.1.1.attn.sr.bias, backbone.layers.1.1.1.attn.norm.weight, backbone.layers.1.1.1.attn.norm.bias, backbone.layers.1.1.1.norm2.weight, backbone.layers.1.1.1.norm2.bias, backbone.layers.1.1.1.ffn.layers.0.weight, backbone.layers.1.1.1.ffn.layers.0.bias, backbone.layers.1.1.1.ffn.layers.1.weight, backbone.layers.1.1.1.ffn.layers.1.bias, backbone.layers.1.1.1.ffn.layers.4.weight, backbone.layers.1.1.1.ffn.layers.4.bias, backbone.layers.1.1.2.norm1.weight, backbone.layers.1.1.2.norm1.bias, backbone.layers.1.1.2.attn.attn.in_proj_weight, backbone.layers.1.1.2.attn.attn.in_proj_bias, backbone.layers.1.1.2.attn.attn.out_proj.weight, backbone.layers.1.1.2.attn.attn.out_proj.bias, backbone.layers.1.1.2.attn.sr.weight, backbone.layers.1.1.2.attn.sr.bias, backbone.layers.1.1.2.attn.norm.weight, backbone.layers.1.1.2.attn.norm.bias, backbone.layers.1.1.2.norm2.weight, backbone.layers.1.1.2.norm2.bias, backbone.layers.1.1.2.ffn.layers.0.weight, backbone.layers.1.1.2.ffn.layers.0.bias, backbone.layers.1.1.2.ffn.layers.1.weight, backbone.layers.1.1.2.ffn.layers.1.bias, backbone.layers.1.1.2.ffn.layers.4.weight, backbone.layers.1.1.2.ffn.layers.4.bias, backbone.layers.1.1.3.norm1.weight, backbone.layers.1.1.3.norm1.bias, backbone.layers.1.1.3.attn.attn.in_proj_weight, backbone.layers.1.1.3.attn.attn.in_proj_bias, backbone.layers.1.1.3.attn.attn.out_proj.weight, backbone.layers.1.1.3.attn.attn.out_proj.bias, backbone.layers.1.1.3.attn.sr.weight, backbone.layers.1.1.3.attn.sr.bias, backbone.layers.1.1.3.attn.norm.weight, backbone.layers.1.1.3.attn.norm.bias, backbone.layers.1.1.3.norm2.weight, backbone.layers.1.1.3.norm2.bias, backbone.layers.1.1.3.ffn.layers.0.weight, backbone.layers.1.1.3.ffn.layers.0.bias, backbone.layers.1.1.3.ffn.layers.1.weight, backbone.layers.1.1.3.ffn.layers.1.bias, backbone.layers.1.1.3.ffn.layers.4.weight, backbone.layers.1.1.3.ffn.layers.4.bias, backbone.layers.1.2.weight, backbone.layers.1.2.bias, backbone.layers.2.0.projection.weight, backbone.layers.2.0.projection.bias, backbone.layers.2.0.norm.weight, backbone.layers.2.0.norm.bias, backbone.layers.2.1.0.norm1.weight, backbone.layers.2.1.0.norm1.bias, backbone.layers.2.1.0.attn.attn.in_proj_weight, backbone.layers.2.1.0.attn.attn.in_proj_bias, backbone.layers.2.1.0.attn.attn.out_proj.weight, backbone.layers.2.1.0.attn.attn.out_proj.bias, backbone.layers.2.1.0.attn.sr.weight, backbone.layers.2.1.0.attn.sr.bias, backbone.layers.2.1.0.attn.norm.weight, backbone.layers.2.1.0.attn.norm.bias, backbone.layers.2.1.0.norm2.weight, backbone.layers.2.1.0.norm2.bias, backbone.layers.2.1.0.ffn.layers.0.weight, backbone.layers.2.1.0.ffn.layers.0.bias, backbone.layers.2.1.0.ffn.layers.1.weight, backbone.layers.2.1.0.ffn.layers.1.bias, backbone.layers.2.1.0.ffn.layers.4.weight, backbone.layers.2.1.0.ffn.layers.4.bias, backbone.layers.2.1.1.norm1.weight, backbone.layers.2.1.1.norm1.bias, backbone.layers.2.1.1.attn.attn.in_proj_weight, backbone.layers.2.1.1.attn.attn.in_proj_bias, backbone.layers.2.1.1.attn.attn.out_proj.weight, backbone.layers.2.1.1.attn.attn.out_proj.bias, backbone.layers.2.1.1.attn.sr.weight, backbone.layers.2.1.1.attn.sr.bias, backbone.layers.2.1.1.attn.norm.weight, backbone.layers.2.1.1.attn.norm.bias, backbone.layers.2.1.1.norm2.weight, backbone.layers.2.1.1.norm2.bias, backbone.layers.2.1.1.ffn.layers.0.weight, backbone.layers.2.1.1.ffn.layers.0.bias, backbone.layers.2.1.1.ffn.layers.1.weight, backbone.layers.2.1.1.ffn.layers.1.bias, backbone.layers.2.1.1.ffn.layers.4.weight, backbone.layers.2.1.1.ffn.layers.4.bias, backbone.layers.2.1.2.norm1.weight, backbone.layers.2.1.2.norm1.bias, backbone.layers.2.1.2.attn.attn.in_proj_weight, backbone.layers.2.1.2.attn.attn.in_proj_bias, backbone.layers.2.1.2.attn.attn.out_proj.weight, backbone.layers.2.1.2.attn.attn.out_proj.bias, backbone.layers.2.1.2.attn.sr.weight, backbone.layers.2.1.2.attn.sr.bias, backbone.layers.2.1.2.attn.norm.weight, backbone.layers.2.1.2.attn.norm.bias, backbone.layers.2.1.2.norm2.weight, backbone.layers.2.1.2.norm2.bias, backbone.layers.2.1.2.ffn.layers.0.weight, backbone.layers.2.1.2.ffn.layers.0.bias, backbone.layers.2.1.2.ffn.layers.1.weight, backbone.layers.2.1.2.ffn.layers.1.bias, backbone.layers.2.1.2.ffn.layers.4.weight, backbone.layers.2.1.2.ffn.layers.4.bias, backbone.layers.2.1.3.norm1.weight, backbone.layers.2.1.3.norm1.bias, backbone.layers.2.1.3.attn.attn.in_proj_weight, backbone.layers.2.1.3.attn.attn.in_proj_bias, backbone.layers.2.1.3.attn.attn.out_proj.weight, backbone.layers.2.1.3.attn.attn.out_proj.bias, backbone.layers.2.1.3.attn.sr.weight, backbone.layers.2.1.3.attn.sr.bias, backbone.layers.2.1.3.attn.norm.weight, backbone.layers.2.1.3.attn.norm.bias, backbone.layers.2.1.3.norm2.weight, backbone.layers.2.1.3.norm2.bias, backbone.layers.2.1.3.ffn.layers.0.weight, backbone.layers.2.1.3.ffn.layers.0.bias, backbone.layers.2.1.3.ffn.layers.1.weight, backbone.layers.2.1.3.ffn.layers.1.bias, backbone.layers.2.1.3.ffn.layers.4.weight, backbone.layers.2.1.3.ffn.layers.4.bias, backbone.layers.2.1.4.norm1.weight, backbone.layers.2.1.4.norm1.bias, backbone.layers.2.1.4.attn.attn.in_proj_weight, backbone.layers.2.1.4.attn.attn.in_proj_bias, backbone.layers.2.1.4.attn.attn.out_proj.weight, backbone.layers.2.1.4.attn.attn.out_proj.bias, backbone.layers.2.1.4.attn.sr.weight, backbone.layers.2.1.4.attn.sr.bias, backbone.layers.2.1.4.attn.norm.weight, backbone.layers.2.1.4.attn.norm.bias, backbone.layers.2.1.4.norm2.weight, backbone.layers.2.1.4.norm2.bias, backbone.layers.2.1.4.ffn.layers.0.weight, backbone.layers.2.1.4.ffn.layers.0.bias, backbone.layers.2.1.4.ffn.layers.1.weight, backbone.layers.2.1.4.ffn.layers.1.bias, backbone.layers.2.1.4.ffn.layers.4.weight, backbone.layers.2.1.4.ffn.layers.4.bias, backbone.layers.2.1.5.norm1.weight, backbone.layers.2.1.5.norm1.bias, backbone.layers.2.1.5.attn.attn.in_proj_weight, backbone.layers.2.1.5.attn.attn.in_proj_bias, backbone.layers.2.1.5.attn.attn.out_proj.weight, backbone.layers.2.1.5.attn.attn.out_proj.bias, backbone.layers.2.1.5.attn.sr.weight, backbone.layers.2.1.5.attn.sr.bias, backbone.layers.2.1.5.attn.norm.weight, backbone.layers.2.1.5.attn.norm.bias, backbone.layers.2.1.5.norm2.weight, backbone.layers.2.1.5.norm2.bias, backbone.layers.2.1.5.ffn.layers.0.weight, backbone.layers.2.1.5.ffn.layers.0.bias, backbone.layers.2.1.5.ffn.layers.1.weight, backbone.layers.2.1.5.ffn.layers.1.bias, backbone.layers.2.1.5.ffn.layers.4.weight, backbone.layers.2.1.5.ffn.layers.4.bias, backbone.layers.2.2.weight, backbone.layers.2.2.bias, backbone.layers.3.0.projection.weight, backbone.layers.3.0.projection.bias, backbone.layers.3.0.norm.weight, backbone.layers.3.0.norm.bias, backbone.layers.3.1.0.norm1.weight, backbone.layers.3.1.0.norm1.bias, backbone.layers.3.1.0.attn.attn.in_proj_weight, backbone.layers.3.1.0.attn.attn.in_proj_bias, backbone.layers.3.1.0.attn.attn.out_proj.weight, backbone.layers.3.1.0.attn.attn.out_proj.bias, backbone.layers.3.1.0.norm2.weight, backbone.layers.3.1.0.norm2.bias, backbone.layers.3.1.0.ffn.layers.0.weight, backbone.layers.3.1.0.ffn.layers.0.bias, backbone.layers.3.1.0.ffn.layers.1.weight, backbone.layers.3.1.0.ffn.layers.1.bias, backbone.layers.3.1.0.ffn.layers.4.weight, backbone.layers.3.1.0.ffn.layers.4.bias, backbone.layers.3.1.1.norm1.weight, backbone.layers.3.1.1.norm1.bias, backbone.layers.3.1.1.attn.attn.in_proj_weight, backbone.layers.3.1.1.attn.attn.in_proj_bias, backbone.layers.3.1.1.attn.attn.out_proj.weight, backbone.layers.3.1.1.attn.attn.out_proj.bias, backbone.layers.3.1.1.norm2.weight, backbone.layers.3.1.1.norm2.bias, backbone.layers.3.1.1.ffn.layers.0.weight, backbone.layers.3.1.1.ffn.layers.0.bias, backbone.layers.3.1.1.ffn.layers.1.weight, backbone.layers.3.1.1.ffn.layers.1.bias, backbone.layers.3.1.1.ffn.layers.4.weight, backbone.layers.3.1.1.ffn.layers.4.bias, backbone.layers.3.1.2.norm1.weight, backbone.layers.3.1.2.norm1.bias, backbone.layers.3.1.2.attn.attn.in_proj_weight, backbone.layers.3.1.2.attn.attn.in_proj_bias, backbone.layers.3.1.2.attn.attn.out_proj.weight, backbone.layers.3.1.2.attn.attn.out_proj.bias, backbone.layers.3.1.2.norm2.weight, backbone.layers.3.1.2.norm2.bias, backbone.layers.3.1.2.ffn.layers.0.weight, backbone.layers.3.1.2.ffn.layers.0.bias, backbone.layers.3.1.2.ffn.layers.1.weight, backbone.layers.3.1.2.ffn.layers.1.bias, backbone.layers.3.1.2.ffn.layers.4.weight, backbone.layers.3.1.2.ffn.layers.4.bias, backbone.layers.3.2.weight, backbone.layers.3.2.bias
+
+missing keys in source state_dict: unet.init_conv.weight, unet.init_conv.bias, unet.time_mlp.1.weight, unet.time_mlp.1.bias, unet.time_mlp.3.weight, unet.time_mlp.3.bias, unet.downs.0.0.mlp.1.weight, unet.downs.0.0.mlp.1.bias, unet.downs.0.0.block1.proj.weight, unet.downs.0.0.block1.proj.bias, unet.downs.0.0.block1.norm.weight, unet.downs.0.0.block1.norm.bias, unet.downs.0.0.block2.proj.weight, unet.downs.0.0.block2.proj.bias, unet.downs.0.0.block2.norm.weight, unet.downs.0.0.block2.norm.bias, unet.downs.0.1.mlp.1.weight, unet.downs.0.1.mlp.1.bias, unet.downs.0.1.block1.proj.weight, unet.downs.0.1.block1.proj.bias, unet.downs.0.1.block1.norm.weight, unet.downs.0.1.block1.norm.bias, unet.downs.0.1.block2.proj.weight, unet.downs.0.1.block2.proj.bias, unet.downs.0.1.block2.norm.weight, unet.downs.0.1.block2.norm.bias, unet.downs.0.2.fn.fn.to_qkv.weight, unet.downs.0.2.fn.fn.to_out.0.weight, unet.downs.0.2.fn.fn.to_out.0.bias, unet.downs.0.2.fn.fn.to_out.1.g, unet.downs.0.2.fn.norm.g, unet.downs.0.3.weight, unet.downs.0.3.bias, unet.downs.1.0.mlp.1.weight, unet.downs.1.0.mlp.1.bias, unet.downs.1.0.block1.proj.weight, unet.downs.1.0.block1.proj.bias, unet.downs.1.0.block1.norm.weight, unet.downs.1.0.block1.norm.bias, unet.downs.1.0.block2.proj.weight, unet.downs.1.0.block2.proj.bias, unet.downs.1.0.block2.norm.weight, unet.downs.1.0.block2.norm.bias, unet.downs.1.1.mlp.1.weight, unet.downs.1.1.mlp.1.bias, unet.downs.1.1.block1.proj.weight, unet.downs.1.1.block1.proj.bias, unet.downs.1.1.block1.norm.weight, unet.downs.1.1.block1.norm.bias, unet.downs.1.1.block2.proj.weight, unet.downs.1.1.block2.proj.bias, unet.downs.1.1.block2.norm.weight, unet.downs.1.1.block2.norm.bias, unet.downs.1.2.fn.fn.to_qkv.weight, unet.downs.1.2.fn.fn.to_out.0.weight, unet.downs.1.2.fn.fn.to_out.0.bias, unet.downs.1.2.fn.fn.to_out.1.g, unet.downs.1.2.fn.norm.g, unet.downs.1.3.weight, unet.downs.1.3.bias, unet.downs.2.0.mlp.1.weight, unet.downs.2.0.mlp.1.bias, unet.downs.2.0.block1.proj.weight, unet.downs.2.0.block1.proj.bias, unet.downs.2.0.block1.norm.weight, unet.downs.2.0.block1.norm.bias, unet.downs.2.0.block2.proj.weight, unet.downs.2.0.block2.proj.bias, unet.downs.2.0.block2.norm.weight, unet.downs.2.0.block2.norm.bias, unet.downs.2.1.mlp.1.weight, unet.downs.2.1.mlp.1.bias, unet.downs.2.1.block1.proj.weight, unet.downs.2.1.block1.proj.bias, unet.downs.2.1.block1.norm.weight, unet.downs.2.1.block1.norm.bias, unet.downs.2.1.block2.proj.weight, unet.downs.2.1.block2.proj.bias, unet.downs.2.1.block2.norm.weight, unet.downs.2.1.block2.norm.bias, unet.downs.2.2.fn.fn.to_qkv.weight, unet.downs.2.2.fn.fn.to_out.0.weight, unet.downs.2.2.fn.fn.to_out.0.bias, unet.downs.2.2.fn.fn.to_out.1.g, unet.downs.2.2.fn.norm.g, unet.downs.2.3.weight, unet.downs.2.3.bias, unet.ups.0.0.mlp.1.weight, unet.ups.0.0.mlp.1.bias, unet.ups.0.0.block1.proj.weight, unet.ups.0.0.block1.proj.bias, unet.ups.0.0.block1.norm.weight, unet.ups.0.0.block1.norm.bias, unet.ups.0.0.block2.proj.weight, unet.ups.0.0.block2.proj.bias, unet.ups.0.0.block2.norm.weight, unet.ups.0.0.block2.norm.bias, unet.ups.0.0.res_conv.weight, unet.ups.0.0.res_conv.bias, unet.ups.0.1.mlp.1.weight, unet.ups.0.1.mlp.1.bias, unet.ups.0.1.block1.proj.weight, unet.ups.0.1.block1.proj.bias, unet.ups.0.1.block1.norm.weight, unet.ups.0.1.block1.norm.bias, unet.ups.0.1.block2.proj.weight, unet.ups.0.1.block2.proj.bias, unet.ups.0.1.block2.norm.weight, unet.ups.0.1.block2.norm.bias, unet.ups.0.1.res_conv.weight, unet.ups.0.1.res_conv.bias, unet.ups.0.2.fn.fn.to_qkv.weight, unet.ups.0.2.fn.fn.to_out.0.weight, unet.ups.0.2.fn.fn.to_out.0.bias, unet.ups.0.2.fn.fn.to_out.1.g, unet.ups.0.2.fn.norm.g, unet.ups.0.3.1.weight, unet.ups.0.3.1.bias, unet.ups.1.0.mlp.1.weight, unet.ups.1.0.mlp.1.bias, unet.ups.1.0.block1.proj.weight, unet.ups.1.0.block1.proj.bias, unet.ups.1.0.block1.norm.weight, unet.ups.1.0.block1.norm.bias, unet.ups.1.0.block2.proj.weight, unet.ups.1.0.block2.proj.bias, unet.ups.1.0.block2.norm.weight, unet.ups.1.0.block2.norm.bias, unet.ups.1.0.res_conv.weight, unet.ups.1.0.res_conv.bias, unet.ups.1.1.mlp.1.weight, unet.ups.1.1.mlp.1.bias, unet.ups.1.1.block1.proj.weight, unet.ups.1.1.block1.proj.bias, unet.ups.1.1.block1.norm.weight, unet.ups.1.1.block1.norm.bias, unet.ups.1.1.block2.proj.weight, unet.ups.1.1.block2.proj.bias, unet.ups.1.1.block2.norm.weight, unet.ups.1.1.block2.norm.bias, unet.ups.1.1.res_conv.weight, unet.ups.1.1.res_conv.bias, unet.ups.1.2.fn.fn.to_qkv.weight, unet.ups.1.2.fn.fn.to_out.0.weight, unet.ups.1.2.fn.fn.to_out.0.bias, unet.ups.1.2.fn.fn.to_out.1.g, unet.ups.1.2.fn.norm.g, unet.ups.1.3.1.weight, unet.ups.1.3.1.bias, unet.ups.2.0.mlp.1.weight, unet.ups.2.0.mlp.1.bias, unet.ups.2.0.block1.proj.weight, unet.ups.2.0.block1.proj.bias, unet.ups.2.0.block1.norm.weight, unet.ups.2.0.block1.norm.bias, unet.ups.2.0.block2.proj.weight, unet.ups.2.0.block2.proj.bias, unet.ups.2.0.block2.norm.weight, unet.ups.2.0.block2.norm.bias, unet.ups.2.0.res_conv.weight, unet.ups.2.0.res_conv.bias, unet.ups.2.1.mlp.1.weight, unet.ups.2.1.mlp.1.bias, unet.ups.2.1.block1.proj.weight, unet.ups.2.1.block1.proj.bias, unet.ups.2.1.block1.norm.weight, unet.ups.2.1.block1.norm.bias, unet.ups.2.1.block2.proj.weight, unet.ups.2.1.block2.proj.bias, unet.ups.2.1.block2.norm.weight, unet.ups.2.1.block2.norm.bias, unet.ups.2.1.res_conv.weight, unet.ups.2.1.res_conv.bias, unet.ups.2.2.fn.fn.to_qkv.weight, unet.ups.2.2.fn.fn.to_out.0.weight, unet.ups.2.2.fn.fn.to_out.0.bias, unet.ups.2.2.fn.fn.to_out.1.g, unet.ups.2.2.fn.norm.g, unet.ups.2.3.weight, unet.ups.2.3.bias, unet.mid_block1.mlp.1.weight, unet.mid_block1.mlp.1.bias, unet.mid_block1.block1.proj.weight, unet.mid_block1.block1.proj.bias, unet.mid_block1.block1.norm.weight, unet.mid_block1.block1.norm.bias, unet.mid_block1.block2.proj.weight, unet.mid_block1.block2.proj.bias, unet.mid_block1.block2.norm.weight, unet.mid_block1.block2.norm.bias, unet.mid_attn.fn.fn.to_qkv.weight, unet.mid_attn.fn.fn.to_out.weight, unet.mid_attn.fn.fn.to_out.bias, unet.mid_attn.fn.norm.g, unet.mid_block2.mlp.1.weight, unet.mid_block2.mlp.1.bias, unet.mid_block2.block1.proj.weight, unet.mid_block2.block1.proj.bias, unet.mid_block2.block1.norm.weight, unet.mid_block2.block1.norm.bias, unet.mid_block2.block2.proj.weight, unet.mid_block2.block2.proj.bias, unet.mid_block2.block2.norm.weight, unet.mid_block2.block2.norm.bias, unet.final_res_block.mlp.1.weight, unet.final_res_block.mlp.1.bias, unet.final_res_block.block1.proj.weight, unet.final_res_block.block1.proj.bias, unet.final_res_block.block1.norm.weight, unet.final_res_block.block1.norm.bias, unet.final_res_block.block2.proj.weight, unet.final_res_block.block2.proj.bias, unet.final_res_block.block2.norm.weight, unet.final_res_block.block2.norm.bias, unet.final_res_block.res_conv.weight, unet.final_res_block.res_conv.bias, unet.final_conv.weight, unet.final_conv.bias, conv_seg_new.weight, conv_seg_new.bias, embed.weight
+
+2023-03-04 18:46:39,098 - mmseg - INFO - EncoderDecoderFreeze(
+  (backbone): MixVisionTransformerCustomInitWeights(
+    (layers): ModuleList(
+      (0): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(3, 64, kernel_size=(7, 7), stride=(4, 4), padding=(3, 3))
+          (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+      )
+      (1): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+      )
+      (2): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(128, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (4): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (5): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+      )
+      (3): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(320, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+      )
+    )
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+  (decode_head): SegformerHeadUnetFCHeadSingleStepLogits(
+    input_transform=multiple_select, ignore_index=0, align_corners=False
+    (loss_decode): CrossEntropyLoss(avg_non_ignore=False)
+    (conv_seg): Conv2d(256, 150, kernel_size=(1, 1), stride=(1, 1))
+    (dropout): Dropout2d(p=0.1, inplace=False)
+    (convs): ModuleList(
+      (0): ConvModule(
+        (conv): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (1): ConvModule(
+        (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (2): ConvModule(
+        (conv): Conv2d(320, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (3): ConvModule(
+        (conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+    )
+    (fusion_conv): ConvModule(
+      (conv): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+      (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+      (activate): ReLU(inplace=True)
+    )
+    (unet): Unet(
+      (init_conv): Conv2d(166, 128, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3))
+      (time_mlp): Sequential(
+        (0): SinusoidalPosEmb()
+        (1): Linear(in_features=128, out_features=512, bias=True)
+        (2): GELU(approximate='none')
+        (3): Linear(in_features=512, out_features=512, bias=True)
+      )
+      (downs): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (ups): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (mid_block1): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (mid_attn): Residual(
+        (fn): PreNorm(
+          (fn): Attention(
+            (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+            (to_out): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (norm): LayerNorm()
+        )
+      )
+      (mid_block2): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (final_res_block): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+      )
+      (final_conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1))
+    )
+    (conv_seg_new): Conv2d(256, 151, kernel_size=(1, 1), stride=(1, 1))
+    (embed): Embedding(151, 16)
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+)
+2023-03-04 18:46:40,019 - mmseg - INFO - Loaded 20210 images
+2023-03-04 18:46:41,028 - mmseg - INFO - Loaded 2000 images
+2023-03-04 18:46:41,033 - mmseg - INFO - load checkpoint from local path: ./work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/latest.pth
+2023-03-04 18:46:41,696 - mmseg - INFO - resumed from epoch: 13, iter 7999
+2023-03-04 18:46:41,697 - mmseg - INFO - Start running, host: laizeqiang@SH-IDC1-10-140-37-114, work_dir: /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits
+2023-03-04 18:46:41,697 - mmseg - INFO - Hooks will be executed in the following order:
+before_run:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_epoch:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_iter:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+ -------------------- 
+after_train_iter:
+(ABOVE_NORMAL) OptimizerHook                      
+(NORMAL      ) CheckpointHook                     
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_train_epoch:
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_epoch:
+(LOW         ) IterTimerHook                      
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_epoch:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_run:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+2023-03-04 18:46:41,698 - mmseg - INFO - workflow: [('train', 1)], max: 80000 iters
+2023-03-04 18:46:41,698 - mmseg - INFO - Checkpoints will be saved to /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits by HardDiskBackend.

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_184631.log.json

@@ -0,0 +1 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+6749699", "seed": 1082958590, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits.py", "mmseg_version": "0.30.0+6749699", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepLogits',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=166,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 1082958590\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_190322.log

@@ -0,0 +1,1139 @@
+2023-03-04 19:03:22,024 - mmseg - INFO - Multi-processing start method is `None`
+2023-03-04 19:03:22,039 - mmseg - INFO - OpenCV num_threads is `128
+2023-03-04 19:03:22,039 - mmseg - INFO - OMP num threads is 1
+2023-03-04 19:03:22,100 - mmseg - INFO - Environment info:
+------------------------------------------------------------
+sys.platform: linux
+Python: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]
+CUDA available: True
+GPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB
+CUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch
+NVCC: Cuda compilation tools, release 11.6, V11.6.124
+GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)
+PyTorch: 1.13.1
+PyTorch compiling details: PyTorch built with:
+  - GCC 9.3
+  - C++ Version: 201402
+  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
+  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
+  - OpenMP 201511 (a.k.a. OpenMP 4.5)
+  - LAPACK is enabled (usually provided by MKL)
+  - NNPACK is enabled
+  - CPU capability usage: AVX2
+  - CUDA Runtime 11.6
+  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
+  - CuDNN 8.3.2  (built against CUDA 11.5)
+  - Magma 2.6.1
+  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 
+
+TorchVision: 0.14.1
+OpenCV: 4.7.0
+MMCV: 1.7.1
+MMCV Compiler: GCC 9.3
+MMCV CUDA Compiler: 11.6
+MMSegmentation: 0.30.0+6749699
+------------------------------------------------------------
+
+2023-03-04 19:03:22,100 - mmseg - INFO - Distributed training: True
+2023-03-04 19:03:22,820 - mmseg - INFO - Config:
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepLogits',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=166,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'
+gpu_ids = range(0, 8)
+auto_resume = True
+
+2023-03-04 19:03:27,162 - mmseg - INFO - Set random seed to 1480177113, deterministic: False
+2023-03-04 19:03:27,413 - mmseg - INFO - Parameters in backbone freezed!
+2023-03-04 19:03:27,414 - mmseg - INFO - Trainable parameters in SegformerHeadUnetFCHeadSingleStep: ['unet.init_conv.weight', 'unet.init_conv.bias', 'unet.time_mlp.1.weight', 'unet.time_mlp.1.bias', 'unet.time_mlp.3.weight', 'unet.time_mlp.3.bias', 'unet.downs.0.0.mlp.1.weight', 'unet.downs.0.0.mlp.1.bias', 'unet.downs.0.0.block1.proj.weight', 'unet.downs.0.0.block1.proj.bias', 'unet.downs.0.0.block1.norm.weight', 'unet.downs.0.0.block1.norm.bias', 'unet.downs.0.0.block2.proj.weight', 'unet.downs.0.0.block2.proj.bias', 'unet.downs.0.0.block2.norm.weight', 'unet.downs.0.0.block2.norm.bias', 'unet.downs.0.1.mlp.1.weight', 'unet.downs.0.1.mlp.1.bias', 'unet.downs.0.1.block1.proj.weight', 'unet.downs.0.1.block1.proj.bias', 'unet.downs.0.1.block1.norm.weight', 'unet.downs.0.1.block1.norm.bias', 'unet.downs.0.1.block2.proj.weight', 'unet.downs.0.1.block2.proj.bias', 'unet.downs.0.1.block2.norm.weight', 'unet.downs.0.1.block2.norm.bias', 'unet.downs.0.2.fn.fn.to_qkv.weight', 'unet.downs.0.2.fn.fn.to_out.0.weight', 'unet.downs.0.2.fn.fn.to_out.0.bias', 'unet.downs.0.2.fn.fn.to_out.1.g', 'unet.downs.0.2.fn.norm.g', 'unet.downs.0.3.weight', 'unet.downs.0.3.bias', 'unet.downs.1.0.mlp.1.weight', 'unet.downs.1.0.mlp.1.bias', 'unet.downs.1.0.block1.proj.weight', 'unet.downs.1.0.block1.proj.bias', 'unet.downs.1.0.block1.norm.weight', 'unet.downs.1.0.block1.norm.bias', 'unet.downs.1.0.block2.proj.weight', 'unet.downs.1.0.block2.proj.bias', 'unet.downs.1.0.block2.norm.weight', 'unet.downs.1.0.block2.norm.bias', 'unet.downs.1.1.mlp.1.weight', 'unet.downs.1.1.mlp.1.bias', 'unet.downs.1.1.block1.proj.weight', 'unet.downs.1.1.block1.proj.bias', 'unet.downs.1.1.block1.norm.weight', 'unet.downs.1.1.block1.norm.bias', 'unet.downs.1.1.block2.proj.weight', 'unet.downs.1.1.block2.proj.bias', 'unet.downs.1.1.block2.norm.weight', 'unet.downs.1.1.block2.norm.bias', 'unet.downs.1.2.fn.fn.to_qkv.weight', 'unet.downs.1.2.fn.fn.to_out.0.weight', 'unet.downs.1.2.fn.fn.to_out.0.bias', 'unet.downs.1.2.fn.fn.to_out.1.g', 'unet.downs.1.2.fn.norm.g', 'unet.downs.1.3.weight', 'unet.downs.1.3.bias', 'unet.downs.2.0.mlp.1.weight', 'unet.downs.2.0.mlp.1.bias', 'unet.downs.2.0.block1.proj.weight', 'unet.downs.2.0.block1.proj.bias', 'unet.downs.2.0.block1.norm.weight', 'unet.downs.2.0.block1.norm.bias', 'unet.downs.2.0.block2.proj.weight', 'unet.downs.2.0.block2.proj.bias', 'unet.downs.2.0.block2.norm.weight', 'unet.downs.2.0.block2.norm.bias', 'unet.downs.2.1.mlp.1.weight', 'unet.downs.2.1.mlp.1.bias', 'unet.downs.2.1.block1.proj.weight', 'unet.downs.2.1.block1.proj.bias', 'unet.downs.2.1.block1.norm.weight', 'unet.downs.2.1.block1.norm.bias', 'unet.downs.2.1.block2.proj.weight', 'unet.downs.2.1.block2.proj.bias', 'unet.downs.2.1.block2.norm.weight', 'unet.downs.2.1.block2.norm.bias', 'unet.downs.2.2.fn.fn.to_qkv.weight', 'unet.downs.2.2.fn.fn.to_out.0.weight', 'unet.downs.2.2.fn.fn.to_out.0.bias', 'unet.downs.2.2.fn.fn.to_out.1.g', 'unet.downs.2.2.fn.norm.g', 'unet.downs.2.3.weight', 'unet.downs.2.3.bias', 'unet.ups.0.0.mlp.1.weight', 'unet.ups.0.0.mlp.1.bias', 'unet.ups.0.0.block1.proj.weight', 'unet.ups.0.0.block1.proj.bias', 'unet.ups.0.0.block1.norm.weight', 'unet.ups.0.0.block1.norm.bias', 'unet.ups.0.0.block2.proj.weight', 'unet.ups.0.0.block2.proj.bias', 'unet.ups.0.0.block2.norm.weight', 'unet.ups.0.0.block2.norm.bias', 'unet.ups.0.0.res_conv.weight', 'unet.ups.0.0.res_conv.bias', 'unet.ups.0.1.mlp.1.weight', 'unet.ups.0.1.mlp.1.bias', 'unet.ups.0.1.block1.proj.weight', 'unet.ups.0.1.block1.proj.bias', 'unet.ups.0.1.block1.norm.weight', 'unet.ups.0.1.block1.norm.bias', 'unet.ups.0.1.block2.proj.weight', 'unet.ups.0.1.block2.proj.bias', 'unet.ups.0.1.block2.norm.weight', 'unet.ups.0.1.block2.norm.bias', 'unet.ups.0.1.res_conv.weight', 'unet.ups.0.1.res_conv.bias', 'unet.ups.0.2.fn.fn.to_qkv.weight', 'unet.ups.0.2.fn.fn.to_out.0.weight', 'unet.ups.0.2.fn.fn.to_out.0.bias', 'unet.ups.0.2.fn.fn.to_out.1.g', 'unet.ups.0.2.fn.norm.g', 'unet.ups.0.3.1.weight', 'unet.ups.0.3.1.bias', 'unet.ups.1.0.mlp.1.weight', 'unet.ups.1.0.mlp.1.bias', 'unet.ups.1.0.block1.proj.weight', 'unet.ups.1.0.block1.proj.bias', 'unet.ups.1.0.block1.norm.weight', 'unet.ups.1.0.block1.norm.bias', 'unet.ups.1.0.block2.proj.weight', 'unet.ups.1.0.block2.proj.bias', 'unet.ups.1.0.block2.norm.weight', 'unet.ups.1.0.block2.norm.bias', 'unet.ups.1.0.res_conv.weight', 'unet.ups.1.0.res_conv.bias', 'unet.ups.1.1.mlp.1.weight', 'unet.ups.1.1.mlp.1.bias', 'unet.ups.1.1.block1.proj.weight', 'unet.ups.1.1.block1.proj.bias', 'unet.ups.1.1.block1.norm.weight', 'unet.ups.1.1.block1.norm.bias', 'unet.ups.1.1.block2.proj.weight', 'unet.ups.1.1.block2.proj.bias', 'unet.ups.1.1.block2.norm.weight', 'unet.ups.1.1.block2.norm.bias', 'unet.ups.1.1.res_conv.weight', 'unet.ups.1.1.res_conv.bias', 'unet.ups.1.2.fn.fn.to_qkv.weight', 'unet.ups.1.2.fn.fn.to_out.0.weight', 'unet.ups.1.2.fn.fn.to_out.0.bias', 'unet.ups.1.2.fn.fn.to_out.1.g', 'unet.ups.1.2.fn.norm.g', 'unet.ups.1.3.1.weight', 'unet.ups.1.3.1.bias', 'unet.ups.2.0.mlp.1.weight', 'unet.ups.2.0.mlp.1.bias', 'unet.ups.2.0.block1.proj.weight', 'unet.ups.2.0.block1.proj.bias', 'unet.ups.2.0.block1.norm.weight', 'unet.ups.2.0.block1.norm.bias', 'unet.ups.2.0.block2.proj.weight', 'unet.ups.2.0.block2.proj.bias', 'unet.ups.2.0.block2.norm.weight', 'unet.ups.2.0.block2.norm.bias', 'unet.ups.2.0.res_conv.weight', 'unet.ups.2.0.res_conv.bias', 'unet.ups.2.1.mlp.1.weight', 'unet.ups.2.1.mlp.1.bias', 'unet.ups.2.1.block1.proj.weight', 'unet.ups.2.1.block1.proj.bias', 'unet.ups.2.1.block1.norm.weight', 'unet.ups.2.1.block1.norm.bias', 'unet.ups.2.1.block2.proj.weight', 'unet.ups.2.1.block2.proj.bias', 'unet.ups.2.1.block2.norm.weight', 'unet.ups.2.1.block2.norm.bias', 'unet.ups.2.1.res_conv.weight', 'unet.ups.2.1.res_conv.bias', 'unet.ups.2.2.fn.fn.to_qkv.weight', 'unet.ups.2.2.fn.fn.to_out.0.weight', 'unet.ups.2.2.fn.fn.to_out.0.bias', 'unet.ups.2.2.fn.fn.to_out.1.g', 'unet.ups.2.2.fn.norm.g', 'unet.ups.2.3.weight', 'unet.ups.2.3.bias', 'unet.mid_block1.mlp.1.weight', 'unet.mid_block1.mlp.1.bias', 'unet.mid_block1.block1.proj.weight', 'unet.mid_block1.block1.proj.bias', 'unet.mid_block1.block1.norm.weight', 'unet.mid_block1.block1.norm.bias', 'unet.mid_block1.block2.proj.weight', 'unet.mid_block1.block2.proj.bias', 'unet.mid_block1.block2.norm.weight', 'unet.mid_block1.block2.norm.bias', 'unet.mid_attn.fn.fn.to_qkv.weight', 'unet.mid_attn.fn.fn.to_out.weight', 'unet.mid_attn.fn.fn.to_out.bias', 'unet.mid_attn.fn.norm.g', 'unet.mid_block2.mlp.1.weight', 'unet.mid_block2.mlp.1.bias', 'unet.mid_block2.block1.proj.weight', 'unet.mid_block2.block1.proj.bias', 'unet.mid_block2.block1.norm.weight', 'unet.mid_block2.block1.norm.bias', 'unet.mid_block2.block2.proj.weight', 'unet.mid_block2.block2.proj.bias', 'unet.mid_block2.block2.norm.weight', 'unet.mid_block2.block2.norm.bias', 'unet.final_res_block.mlp.1.weight', 'unet.final_res_block.mlp.1.bias', 'unet.final_res_block.block1.proj.weight', 'unet.final_res_block.block1.proj.bias', 'unet.final_res_block.block1.norm.weight', 'unet.final_res_block.block1.norm.bias', 'unet.final_res_block.block2.proj.weight', 'unet.final_res_block.block2.proj.bias', 'unet.final_res_block.block2.norm.weight', 'unet.final_res_block.block2.norm.bias', 'unet.final_res_block.res_conv.weight', 'unet.final_res_block.res_conv.bias', 'unet.final_conv.weight', 'unet.final_conv.bias', 'conv_seg_new.weight', 'conv_seg_new.bias']
+2023-03-04 19:03:27,414 - mmseg - INFO - Parameters in decode_head freezed!
+2023-03-04 19:03:27,436 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 19:03:27,682 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: decode_head.conv_seg.weight, decode_head.conv_seg.bias, decode_head.convs.0.conv.weight, decode_head.convs.0.bn.weight, decode_head.convs.0.bn.bias, decode_head.convs.0.bn.running_mean, decode_head.convs.0.bn.running_var, decode_head.convs.0.bn.num_batches_tracked, decode_head.convs.1.conv.weight, decode_head.convs.1.bn.weight, decode_head.convs.1.bn.bias, decode_head.convs.1.bn.running_mean, decode_head.convs.1.bn.running_var, decode_head.convs.1.bn.num_batches_tracked, decode_head.convs.2.conv.weight, decode_head.convs.2.bn.weight, decode_head.convs.2.bn.bias, decode_head.convs.2.bn.running_mean, decode_head.convs.2.bn.running_var, decode_head.convs.2.bn.num_batches_tracked, decode_head.convs.3.conv.weight, decode_head.convs.3.bn.weight, decode_head.convs.3.bn.bias, decode_head.convs.3.bn.running_mean, decode_head.convs.3.bn.running_var, decode_head.convs.3.bn.num_batches_tracked, decode_head.fusion_conv.conv.weight, decode_head.fusion_conv.bn.weight, decode_head.fusion_conv.bn.bias, decode_head.fusion_conv.bn.running_mean, decode_head.fusion_conv.bn.running_var, decode_head.fusion_conv.bn.num_batches_tracked
+
+2023-03-04 19:03:27,695 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 19:03:27,908 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: backbone.layers.0.0.projection.weight, backbone.layers.0.0.projection.bias, backbone.layers.0.0.norm.weight, backbone.layers.0.0.norm.bias, backbone.layers.0.1.0.norm1.weight, backbone.layers.0.1.0.norm1.bias, backbone.layers.0.1.0.attn.attn.in_proj_weight, backbone.layers.0.1.0.attn.attn.in_proj_bias, backbone.layers.0.1.0.attn.attn.out_proj.weight, backbone.layers.0.1.0.attn.attn.out_proj.bias, backbone.layers.0.1.0.attn.sr.weight, backbone.layers.0.1.0.attn.sr.bias, backbone.layers.0.1.0.attn.norm.weight, backbone.layers.0.1.0.attn.norm.bias, backbone.layers.0.1.0.norm2.weight, backbone.layers.0.1.0.norm2.bias, backbone.layers.0.1.0.ffn.layers.0.weight, backbone.layers.0.1.0.ffn.layers.0.bias, backbone.layers.0.1.0.ffn.layers.1.weight, backbone.layers.0.1.0.ffn.layers.1.bias, backbone.layers.0.1.0.ffn.layers.4.weight, backbone.layers.0.1.0.ffn.layers.4.bias, backbone.layers.0.1.1.norm1.weight, backbone.layers.0.1.1.norm1.bias, backbone.layers.0.1.1.attn.attn.in_proj_weight, backbone.layers.0.1.1.attn.attn.in_proj_bias, backbone.layers.0.1.1.attn.attn.out_proj.weight, backbone.layers.0.1.1.attn.attn.out_proj.bias, backbone.layers.0.1.1.attn.sr.weight, backbone.layers.0.1.1.attn.sr.bias, backbone.layers.0.1.1.attn.norm.weight, backbone.layers.0.1.1.attn.norm.bias, backbone.layers.0.1.1.norm2.weight, backbone.layers.0.1.1.norm2.bias, backbone.layers.0.1.1.ffn.layers.0.weight, backbone.layers.0.1.1.ffn.layers.0.bias, backbone.layers.0.1.1.ffn.layers.1.weight, backbone.layers.0.1.1.ffn.layers.1.bias, backbone.layers.0.1.1.ffn.layers.4.weight, backbone.layers.0.1.1.ffn.layers.4.bias, backbone.layers.0.1.2.norm1.weight, backbone.layers.0.1.2.norm1.bias, backbone.layers.0.1.2.attn.attn.in_proj_weight, backbone.layers.0.1.2.attn.attn.in_proj_bias, backbone.layers.0.1.2.attn.attn.out_proj.weight, backbone.layers.0.1.2.attn.attn.out_proj.bias, backbone.layers.0.1.2.attn.sr.weight, backbone.layers.0.1.2.attn.sr.bias, backbone.layers.0.1.2.attn.norm.weight, backbone.layers.0.1.2.attn.norm.bias, backbone.layers.0.1.2.norm2.weight, backbone.layers.0.1.2.norm2.bias, backbone.layers.0.1.2.ffn.layers.0.weight, backbone.layers.0.1.2.ffn.layers.0.bias, backbone.layers.0.1.2.ffn.layers.1.weight, backbone.layers.0.1.2.ffn.layers.1.bias, backbone.layers.0.1.2.ffn.layers.4.weight, backbone.layers.0.1.2.ffn.layers.4.bias, backbone.layers.0.2.weight, backbone.layers.0.2.bias, backbone.layers.1.0.projection.weight, backbone.layers.1.0.projection.bias, backbone.layers.1.0.norm.weight, backbone.layers.1.0.norm.bias, backbone.layers.1.1.0.norm1.weight, backbone.layers.1.1.0.norm1.bias, backbone.layers.1.1.0.attn.attn.in_proj_weight, backbone.layers.1.1.0.attn.attn.in_proj_bias, backbone.layers.1.1.0.attn.attn.out_proj.weight, backbone.layers.1.1.0.attn.attn.out_proj.bias, backbone.layers.1.1.0.attn.sr.weight, backbone.layers.1.1.0.attn.sr.bias, backbone.layers.1.1.0.attn.norm.weight, backbone.layers.1.1.0.attn.norm.bias, backbone.layers.1.1.0.norm2.weight, backbone.layers.1.1.0.norm2.bias, backbone.layers.1.1.0.ffn.layers.0.weight, backbone.layers.1.1.0.ffn.layers.0.bias, backbone.layers.1.1.0.ffn.layers.1.weight, backbone.layers.1.1.0.ffn.layers.1.bias, backbone.layers.1.1.0.ffn.layers.4.weight, backbone.layers.1.1.0.ffn.layers.4.bias, backbone.layers.1.1.1.norm1.weight, backbone.layers.1.1.1.norm1.bias, backbone.layers.1.1.1.attn.attn.in_proj_weight, backbone.layers.1.1.1.attn.attn.in_proj_bias, backbone.layers.1.1.1.attn.attn.out_proj.weight, backbone.layers.1.1.1.attn.attn.out_proj.bias, backbone.layers.1.1.1.attn.sr.weight, backbone.layers.1.1.1.attn.sr.bias, backbone.layers.1.1.1.attn.norm.weight, backbone.layers.1.1.1.attn.norm.bias, backbone.layers.1.1.1.norm2.weight, backbone.layers.1.1.1.norm2.bias, backbone.layers.1.1.1.ffn.layers.0.weight, backbone.layers.1.1.1.ffn.layers.0.bias, backbone.layers.1.1.1.ffn.layers.1.weight, backbone.layers.1.1.1.ffn.layers.1.bias, backbone.layers.1.1.1.ffn.layers.4.weight, backbone.layers.1.1.1.ffn.layers.4.bias, backbone.layers.1.1.2.norm1.weight, backbone.layers.1.1.2.norm1.bias, backbone.layers.1.1.2.attn.attn.in_proj_weight, backbone.layers.1.1.2.attn.attn.in_proj_bias, backbone.layers.1.1.2.attn.attn.out_proj.weight, backbone.layers.1.1.2.attn.attn.out_proj.bias, backbone.layers.1.1.2.attn.sr.weight, backbone.layers.1.1.2.attn.sr.bias, backbone.layers.1.1.2.attn.norm.weight, backbone.layers.1.1.2.attn.norm.bias, backbone.layers.1.1.2.norm2.weight, backbone.layers.1.1.2.norm2.bias, backbone.layers.1.1.2.ffn.layers.0.weight, backbone.layers.1.1.2.ffn.layers.0.bias, backbone.layers.1.1.2.ffn.layers.1.weight, backbone.layers.1.1.2.ffn.layers.1.bias, backbone.layers.1.1.2.ffn.layers.4.weight, backbone.layers.1.1.2.ffn.layers.4.bias, backbone.layers.1.1.3.norm1.weight, backbone.layers.1.1.3.norm1.bias, backbone.layers.1.1.3.attn.attn.in_proj_weight, backbone.layers.1.1.3.attn.attn.in_proj_bias, backbone.layers.1.1.3.attn.attn.out_proj.weight, backbone.layers.1.1.3.attn.attn.out_proj.bias, backbone.layers.1.1.3.attn.sr.weight, backbone.layers.1.1.3.attn.sr.bias, backbone.layers.1.1.3.attn.norm.weight, backbone.layers.1.1.3.attn.norm.bias, backbone.layers.1.1.3.norm2.weight, backbone.layers.1.1.3.norm2.bias, backbone.layers.1.1.3.ffn.layers.0.weight, backbone.layers.1.1.3.ffn.layers.0.bias, backbone.layers.1.1.3.ffn.layers.1.weight, backbone.layers.1.1.3.ffn.layers.1.bias, backbone.layers.1.1.3.ffn.layers.4.weight, backbone.layers.1.1.3.ffn.layers.4.bias, backbone.layers.1.2.weight, backbone.layers.1.2.bias, backbone.layers.2.0.projection.weight, backbone.layers.2.0.projection.bias, backbone.layers.2.0.norm.weight, backbone.layers.2.0.norm.bias, backbone.layers.2.1.0.norm1.weight, backbone.layers.2.1.0.norm1.bias, backbone.layers.2.1.0.attn.attn.in_proj_weight, backbone.layers.2.1.0.attn.attn.in_proj_bias, backbone.layers.2.1.0.attn.attn.out_proj.weight, backbone.layers.2.1.0.attn.attn.out_proj.bias, backbone.layers.2.1.0.attn.sr.weight, backbone.layers.2.1.0.attn.sr.bias, backbone.layers.2.1.0.attn.norm.weight, backbone.layers.2.1.0.attn.norm.bias, backbone.layers.2.1.0.norm2.weight, backbone.layers.2.1.0.norm2.bias, backbone.layers.2.1.0.ffn.layers.0.weight, backbone.layers.2.1.0.ffn.layers.0.bias, backbone.layers.2.1.0.ffn.layers.1.weight, backbone.layers.2.1.0.ffn.layers.1.bias, backbone.layers.2.1.0.ffn.layers.4.weight, backbone.layers.2.1.0.ffn.layers.4.bias, backbone.layers.2.1.1.norm1.weight, backbone.layers.2.1.1.norm1.bias, backbone.layers.2.1.1.attn.attn.in_proj_weight, backbone.layers.2.1.1.attn.attn.in_proj_bias, backbone.layers.2.1.1.attn.attn.out_proj.weight, backbone.layers.2.1.1.attn.attn.out_proj.bias, backbone.layers.2.1.1.attn.sr.weight, backbone.layers.2.1.1.attn.sr.bias, backbone.layers.2.1.1.attn.norm.weight, backbone.layers.2.1.1.attn.norm.bias, backbone.layers.2.1.1.norm2.weight, backbone.layers.2.1.1.norm2.bias, backbone.layers.2.1.1.ffn.layers.0.weight, backbone.layers.2.1.1.ffn.layers.0.bias, backbone.layers.2.1.1.ffn.layers.1.weight, backbone.layers.2.1.1.ffn.layers.1.bias, backbone.layers.2.1.1.ffn.layers.4.weight, backbone.layers.2.1.1.ffn.layers.4.bias, backbone.layers.2.1.2.norm1.weight, backbone.layers.2.1.2.norm1.bias, backbone.layers.2.1.2.attn.attn.in_proj_weight, backbone.layers.2.1.2.attn.attn.in_proj_bias, backbone.layers.2.1.2.attn.attn.out_proj.weight, backbone.layers.2.1.2.attn.attn.out_proj.bias, backbone.layers.2.1.2.attn.sr.weight, backbone.layers.2.1.2.attn.sr.bias, backbone.layers.2.1.2.attn.norm.weight, backbone.layers.2.1.2.attn.norm.bias, backbone.layers.2.1.2.norm2.weight, backbone.layers.2.1.2.norm2.bias, backbone.layers.2.1.2.ffn.layers.0.weight, backbone.layers.2.1.2.ffn.layers.0.bias, backbone.layers.2.1.2.ffn.layers.1.weight, backbone.layers.2.1.2.ffn.layers.1.bias, backbone.layers.2.1.2.ffn.layers.4.weight, backbone.layers.2.1.2.ffn.layers.4.bias, backbone.layers.2.1.3.norm1.weight, backbone.layers.2.1.3.norm1.bias, backbone.layers.2.1.3.attn.attn.in_proj_weight, backbone.layers.2.1.3.attn.attn.in_proj_bias, backbone.layers.2.1.3.attn.attn.out_proj.weight, backbone.layers.2.1.3.attn.attn.out_proj.bias, backbone.layers.2.1.3.attn.sr.weight, backbone.layers.2.1.3.attn.sr.bias, backbone.layers.2.1.3.attn.norm.weight, backbone.layers.2.1.3.attn.norm.bias, backbone.layers.2.1.3.norm2.weight, backbone.layers.2.1.3.norm2.bias, backbone.layers.2.1.3.ffn.layers.0.weight, backbone.layers.2.1.3.ffn.layers.0.bias, backbone.layers.2.1.3.ffn.layers.1.weight, backbone.layers.2.1.3.ffn.layers.1.bias, backbone.layers.2.1.3.ffn.layers.4.weight, backbone.layers.2.1.3.ffn.layers.4.bias, backbone.layers.2.1.4.norm1.weight, backbone.layers.2.1.4.norm1.bias, backbone.layers.2.1.4.attn.attn.in_proj_weight, backbone.layers.2.1.4.attn.attn.in_proj_bias, backbone.layers.2.1.4.attn.attn.out_proj.weight, backbone.layers.2.1.4.attn.attn.out_proj.bias, backbone.layers.2.1.4.attn.sr.weight, backbone.layers.2.1.4.attn.sr.bias, backbone.layers.2.1.4.attn.norm.weight, backbone.layers.2.1.4.attn.norm.bias, backbone.layers.2.1.4.norm2.weight, backbone.layers.2.1.4.norm2.bias, backbone.layers.2.1.4.ffn.layers.0.weight, backbone.layers.2.1.4.ffn.layers.0.bias, backbone.layers.2.1.4.ffn.layers.1.weight, backbone.layers.2.1.4.ffn.layers.1.bias, backbone.layers.2.1.4.ffn.layers.4.weight, backbone.layers.2.1.4.ffn.layers.4.bias, backbone.layers.2.1.5.norm1.weight, backbone.layers.2.1.5.norm1.bias, backbone.layers.2.1.5.attn.attn.in_proj_weight, backbone.layers.2.1.5.attn.attn.in_proj_bias, backbone.layers.2.1.5.attn.attn.out_proj.weight, backbone.layers.2.1.5.attn.attn.out_proj.bias, backbone.layers.2.1.5.attn.sr.weight, backbone.layers.2.1.5.attn.sr.bias, backbone.layers.2.1.5.attn.norm.weight, backbone.layers.2.1.5.attn.norm.bias, backbone.layers.2.1.5.norm2.weight, backbone.layers.2.1.5.norm2.bias, backbone.layers.2.1.5.ffn.layers.0.weight, backbone.layers.2.1.5.ffn.layers.0.bias, backbone.layers.2.1.5.ffn.layers.1.weight, backbone.layers.2.1.5.ffn.layers.1.bias, backbone.layers.2.1.5.ffn.layers.4.weight, backbone.layers.2.1.5.ffn.layers.4.bias, backbone.layers.2.2.weight, backbone.layers.2.2.bias, backbone.layers.3.0.projection.weight, backbone.layers.3.0.projection.bias, backbone.layers.3.0.norm.weight, backbone.layers.3.0.norm.bias, backbone.layers.3.1.0.norm1.weight, backbone.layers.3.1.0.norm1.bias, backbone.layers.3.1.0.attn.attn.in_proj_weight, backbone.layers.3.1.0.attn.attn.in_proj_bias, backbone.layers.3.1.0.attn.attn.out_proj.weight, backbone.layers.3.1.0.attn.attn.out_proj.bias, backbone.layers.3.1.0.norm2.weight, backbone.layers.3.1.0.norm2.bias, backbone.layers.3.1.0.ffn.layers.0.weight, backbone.layers.3.1.0.ffn.layers.0.bias, backbone.layers.3.1.0.ffn.layers.1.weight, backbone.layers.3.1.0.ffn.layers.1.bias, backbone.layers.3.1.0.ffn.layers.4.weight, backbone.layers.3.1.0.ffn.layers.4.bias, backbone.layers.3.1.1.norm1.weight, backbone.layers.3.1.1.norm1.bias, backbone.layers.3.1.1.attn.attn.in_proj_weight, backbone.layers.3.1.1.attn.attn.in_proj_bias, backbone.layers.3.1.1.attn.attn.out_proj.weight, backbone.layers.3.1.1.attn.attn.out_proj.bias, backbone.layers.3.1.1.norm2.weight, backbone.layers.3.1.1.norm2.bias, backbone.layers.3.1.1.ffn.layers.0.weight, backbone.layers.3.1.1.ffn.layers.0.bias, backbone.layers.3.1.1.ffn.layers.1.weight, backbone.layers.3.1.1.ffn.layers.1.bias, backbone.layers.3.1.1.ffn.layers.4.weight, backbone.layers.3.1.1.ffn.layers.4.bias, backbone.layers.3.1.2.norm1.weight, backbone.layers.3.1.2.norm1.bias, backbone.layers.3.1.2.attn.attn.in_proj_weight, backbone.layers.3.1.2.attn.attn.in_proj_bias, backbone.layers.3.1.2.attn.attn.out_proj.weight, backbone.layers.3.1.2.attn.attn.out_proj.bias, backbone.layers.3.1.2.norm2.weight, backbone.layers.3.1.2.norm2.bias, backbone.layers.3.1.2.ffn.layers.0.weight, backbone.layers.3.1.2.ffn.layers.0.bias, backbone.layers.3.1.2.ffn.layers.1.weight, backbone.layers.3.1.2.ffn.layers.1.bias, backbone.layers.3.1.2.ffn.layers.4.weight, backbone.layers.3.1.2.ffn.layers.4.bias, backbone.layers.3.2.weight, backbone.layers.3.2.bias
+
+missing keys in source state_dict: unet.init_conv.weight, unet.init_conv.bias, unet.time_mlp.1.weight, unet.time_mlp.1.bias, unet.time_mlp.3.weight, unet.time_mlp.3.bias, unet.downs.0.0.mlp.1.weight, unet.downs.0.0.mlp.1.bias, unet.downs.0.0.block1.proj.weight, unet.downs.0.0.block1.proj.bias, unet.downs.0.0.block1.norm.weight, unet.downs.0.0.block1.norm.bias, unet.downs.0.0.block2.proj.weight, unet.downs.0.0.block2.proj.bias, unet.downs.0.0.block2.norm.weight, unet.downs.0.0.block2.norm.bias, unet.downs.0.1.mlp.1.weight, unet.downs.0.1.mlp.1.bias, unet.downs.0.1.block1.proj.weight, unet.downs.0.1.block1.proj.bias, unet.downs.0.1.block1.norm.weight, unet.downs.0.1.block1.norm.bias, unet.downs.0.1.block2.proj.weight, unet.downs.0.1.block2.proj.bias, unet.downs.0.1.block2.norm.weight, unet.downs.0.1.block2.norm.bias, unet.downs.0.2.fn.fn.to_qkv.weight, unet.downs.0.2.fn.fn.to_out.0.weight, unet.downs.0.2.fn.fn.to_out.0.bias, unet.downs.0.2.fn.fn.to_out.1.g, unet.downs.0.2.fn.norm.g, unet.downs.0.3.weight, unet.downs.0.3.bias, unet.downs.1.0.mlp.1.weight, unet.downs.1.0.mlp.1.bias, unet.downs.1.0.block1.proj.weight, unet.downs.1.0.block1.proj.bias, unet.downs.1.0.block1.norm.weight, unet.downs.1.0.block1.norm.bias, unet.downs.1.0.block2.proj.weight, unet.downs.1.0.block2.proj.bias, unet.downs.1.0.block2.norm.weight, unet.downs.1.0.block2.norm.bias, unet.downs.1.1.mlp.1.weight, unet.downs.1.1.mlp.1.bias, unet.downs.1.1.block1.proj.weight, unet.downs.1.1.block1.proj.bias, unet.downs.1.1.block1.norm.weight, unet.downs.1.1.block1.norm.bias, unet.downs.1.1.block2.proj.weight, unet.downs.1.1.block2.proj.bias, unet.downs.1.1.block2.norm.weight, unet.downs.1.1.block2.norm.bias, unet.downs.1.2.fn.fn.to_qkv.weight, unet.downs.1.2.fn.fn.to_out.0.weight, unet.downs.1.2.fn.fn.to_out.0.bias, unet.downs.1.2.fn.fn.to_out.1.g, unet.downs.1.2.fn.norm.g, unet.downs.1.3.weight, unet.downs.1.3.bias, unet.downs.2.0.mlp.1.weight, unet.downs.2.0.mlp.1.bias, unet.downs.2.0.block1.proj.weight, unet.downs.2.0.block1.proj.bias, unet.downs.2.0.block1.norm.weight, unet.downs.2.0.block1.norm.bias, unet.downs.2.0.block2.proj.weight, unet.downs.2.0.block2.proj.bias, unet.downs.2.0.block2.norm.weight, unet.downs.2.0.block2.norm.bias, unet.downs.2.1.mlp.1.weight, unet.downs.2.1.mlp.1.bias, unet.downs.2.1.block1.proj.weight, unet.downs.2.1.block1.proj.bias, unet.downs.2.1.block1.norm.weight, unet.downs.2.1.block1.norm.bias, unet.downs.2.1.block2.proj.weight, unet.downs.2.1.block2.proj.bias, unet.downs.2.1.block2.norm.weight, unet.downs.2.1.block2.norm.bias, unet.downs.2.2.fn.fn.to_qkv.weight, unet.downs.2.2.fn.fn.to_out.0.weight, unet.downs.2.2.fn.fn.to_out.0.bias, unet.downs.2.2.fn.fn.to_out.1.g, unet.downs.2.2.fn.norm.g, unet.downs.2.3.weight, unet.downs.2.3.bias, unet.ups.0.0.mlp.1.weight, unet.ups.0.0.mlp.1.bias, unet.ups.0.0.block1.proj.weight, unet.ups.0.0.block1.proj.bias, unet.ups.0.0.block1.norm.weight, unet.ups.0.0.block1.norm.bias, unet.ups.0.0.block2.proj.weight, unet.ups.0.0.block2.proj.bias, unet.ups.0.0.block2.norm.weight, unet.ups.0.0.block2.norm.bias, unet.ups.0.0.res_conv.weight, unet.ups.0.0.res_conv.bias, unet.ups.0.1.mlp.1.weight, unet.ups.0.1.mlp.1.bias, unet.ups.0.1.block1.proj.weight, unet.ups.0.1.block1.proj.bias, unet.ups.0.1.block1.norm.weight, unet.ups.0.1.block1.norm.bias, unet.ups.0.1.block2.proj.weight, unet.ups.0.1.block2.proj.bias, unet.ups.0.1.block2.norm.weight, unet.ups.0.1.block2.norm.bias, unet.ups.0.1.res_conv.weight, unet.ups.0.1.res_conv.bias, unet.ups.0.2.fn.fn.to_qkv.weight, unet.ups.0.2.fn.fn.to_out.0.weight, unet.ups.0.2.fn.fn.to_out.0.bias, unet.ups.0.2.fn.fn.to_out.1.g, unet.ups.0.2.fn.norm.g, unet.ups.0.3.1.weight, unet.ups.0.3.1.bias, unet.ups.1.0.mlp.1.weight, unet.ups.1.0.mlp.1.bias, unet.ups.1.0.block1.proj.weight, unet.ups.1.0.block1.proj.bias, unet.ups.1.0.block1.norm.weight, unet.ups.1.0.block1.norm.bias, unet.ups.1.0.block2.proj.weight, unet.ups.1.0.block2.proj.bias, unet.ups.1.0.block2.norm.weight, unet.ups.1.0.block2.norm.bias, unet.ups.1.0.res_conv.weight, unet.ups.1.0.res_conv.bias, unet.ups.1.1.mlp.1.weight, unet.ups.1.1.mlp.1.bias, unet.ups.1.1.block1.proj.weight, unet.ups.1.1.block1.proj.bias, unet.ups.1.1.block1.norm.weight, unet.ups.1.1.block1.norm.bias, unet.ups.1.1.block2.proj.weight, unet.ups.1.1.block2.proj.bias, unet.ups.1.1.block2.norm.weight, unet.ups.1.1.block2.norm.bias, unet.ups.1.1.res_conv.weight, unet.ups.1.1.res_conv.bias, unet.ups.1.2.fn.fn.to_qkv.weight, unet.ups.1.2.fn.fn.to_out.0.weight, unet.ups.1.2.fn.fn.to_out.0.bias, unet.ups.1.2.fn.fn.to_out.1.g, unet.ups.1.2.fn.norm.g, unet.ups.1.3.1.weight, unet.ups.1.3.1.bias, unet.ups.2.0.mlp.1.weight, unet.ups.2.0.mlp.1.bias, unet.ups.2.0.block1.proj.weight, unet.ups.2.0.block1.proj.bias, unet.ups.2.0.block1.norm.weight, unet.ups.2.0.block1.norm.bias, unet.ups.2.0.block2.proj.weight, unet.ups.2.0.block2.proj.bias, unet.ups.2.0.block2.norm.weight, unet.ups.2.0.block2.norm.bias, unet.ups.2.0.res_conv.weight, unet.ups.2.0.res_conv.bias, unet.ups.2.1.mlp.1.weight, unet.ups.2.1.mlp.1.bias, unet.ups.2.1.block1.proj.weight, unet.ups.2.1.block1.proj.bias, unet.ups.2.1.block1.norm.weight, unet.ups.2.1.block1.norm.bias, unet.ups.2.1.block2.proj.weight, unet.ups.2.1.block2.proj.bias, unet.ups.2.1.block2.norm.weight, unet.ups.2.1.block2.norm.bias, unet.ups.2.1.res_conv.weight, unet.ups.2.1.res_conv.bias, unet.ups.2.2.fn.fn.to_qkv.weight, unet.ups.2.2.fn.fn.to_out.0.weight, unet.ups.2.2.fn.fn.to_out.0.bias, unet.ups.2.2.fn.fn.to_out.1.g, unet.ups.2.2.fn.norm.g, unet.ups.2.3.weight, unet.ups.2.3.bias, unet.mid_block1.mlp.1.weight, unet.mid_block1.mlp.1.bias, unet.mid_block1.block1.proj.weight, unet.mid_block1.block1.proj.bias, unet.mid_block1.block1.norm.weight, unet.mid_block1.block1.norm.bias, unet.mid_block1.block2.proj.weight, unet.mid_block1.block2.proj.bias, unet.mid_block1.block2.norm.weight, unet.mid_block1.block2.norm.bias, unet.mid_attn.fn.fn.to_qkv.weight, unet.mid_attn.fn.fn.to_out.weight, unet.mid_attn.fn.fn.to_out.bias, unet.mid_attn.fn.norm.g, unet.mid_block2.mlp.1.weight, unet.mid_block2.mlp.1.bias, unet.mid_block2.block1.proj.weight, unet.mid_block2.block1.proj.bias, unet.mid_block2.block1.norm.weight, unet.mid_block2.block1.norm.bias, unet.mid_block2.block2.proj.weight, unet.mid_block2.block2.proj.bias, unet.mid_block2.block2.norm.weight, unet.mid_block2.block2.norm.bias, unet.final_res_block.mlp.1.weight, unet.final_res_block.mlp.1.bias, unet.final_res_block.block1.proj.weight, unet.final_res_block.block1.proj.bias, unet.final_res_block.block1.norm.weight, unet.final_res_block.block1.norm.bias, unet.final_res_block.block2.proj.weight, unet.final_res_block.block2.proj.bias, unet.final_res_block.block2.norm.weight, unet.final_res_block.block2.norm.bias, unet.final_res_block.res_conv.weight, unet.final_res_block.res_conv.bias, unet.final_conv.weight, unet.final_conv.bias, conv_seg_new.weight, conv_seg_new.bias, embed.weight
+
+2023-03-04 19:03:27,934 - mmseg - INFO - EncoderDecoderFreeze(
+  (backbone): MixVisionTransformerCustomInitWeights(
+    (layers): ModuleList(
+      (0): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(3, 64, kernel_size=(7, 7), stride=(4, 4), padding=(3, 3))
+          (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+      )
+      (1): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+      )
+      (2): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(128, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (4): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (5): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+      )
+      (3): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(320, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+      )
+    )
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+  (decode_head): SegformerHeadUnetFCHeadSingleStepLogits(
+    input_transform=multiple_select, ignore_index=0, align_corners=False
+    (loss_decode): CrossEntropyLoss(avg_non_ignore=False)
+    (conv_seg): Conv2d(256, 150, kernel_size=(1, 1), stride=(1, 1))
+    (dropout): Dropout2d(p=0.1, inplace=False)
+    (convs): ModuleList(
+      (0): ConvModule(
+        (conv): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (1): ConvModule(
+        (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (2): ConvModule(
+        (conv): Conv2d(320, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (3): ConvModule(
+        (conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+    )
+    (fusion_conv): ConvModule(
+      (conv): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+      (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+      (activate): ReLU(inplace=True)
+    )
+    (unet): Unet(
+      (init_conv): Conv2d(166, 128, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3))
+      (time_mlp): Sequential(
+        (0): SinusoidalPosEmb()
+        (1): Linear(in_features=128, out_features=512, bias=True)
+        (2): GELU(approximate='none')
+        (3): Linear(in_features=512, out_features=512, bias=True)
+      )
+      (downs): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (ups): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (mid_block1): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (mid_attn): Residual(
+        (fn): PreNorm(
+          (fn): Attention(
+            (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+            (to_out): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (norm): LayerNorm()
+        )
+      )
+      (mid_block2): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (final_res_block): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+      )
+      (final_conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1))
+    )
+    (conv_seg_new): Conv2d(256, 151, kernel_size=(1, 1), stride=(1, 1))
+    (embed): Embedding(151, 16)
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+)
+2023-03-04 19:03:28,858 - mmseg - INFO - Loaded 20210 images
+2023-03-04 19:03:29,858 - mmseg - INFO - Loaded 2000 images
+2023-03-04 19:03:29,859 - mmseg - INFO - load checkpoint from local path: ./work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/latest.pth
+2023-03-04 19:03:30,494 - mmseg - INFO - resumed from epoch: 13, iter 7999
+2023-03-04 19:03:30,496 - mmseg - INFO - Start running, host: laizeqiang@SH-IDC1-10-140-37-114, work_dir: /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits
+2023-03-04 19:03:30,496 - mmseg - INFO - Hooks will be executed in the following order:
+before_run:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_epoch:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_iter:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+ -------------------- 
+after_train_iter:
+(ABOVE_NORMAL) OptimizerHook                      
+(NORMAL      ) CheckpointHook                     
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_train_epoch:
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHook                       
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_epoch:
+(LOW         ) IterTimerHook                      
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_epoch:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_run:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+2023-03-04 19:03:30,496 - mmseg - INFO - workflow: [('train', 1)], max: 80000 iters
+2023-03-04 19:03:30,496 - mmseg - INFO - Checkpoints will be saved to /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits by HardDiskBackend.

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/20230304_190322.log.json

@@ -0,0 +1 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+6749699", "seed": 1480177113, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits.py", "mmseg_version": "0.30.0+6749699", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepLogits',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=166,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 1480177113\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits.py

@@ -0,0 +1,184 @@
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepLogits',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=166,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits'
+gpu_ids = range(0, 8)
+auto_resume = True

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/best_mIoU_iter_72000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd124854833d3fc6ae7e9bebd9cd5f44f52b356d910c3636420abfb59c287ac2
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_16000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f049b01364b5f98b74b11cdd8da2a822c43f907429f22e65833516119b27227
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_24000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bc558779cba0cb559853e558a4e3807455668b4a4b75cae67985c5090052e19
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_32000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:126f85ffef4d1ca315447b2f44f3b2392a6957a63c6d8c7c44f537fe558085c1
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_40000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2a1744562fc230053f45aa4f72a8c66bc1ecfa943ea5c7e3ad0275b6019ca55
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_48000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:513e0b4afde17af815906fb34ba17e59e0b0c5806b3a7afb0a470406f5ea7cb4
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_56000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:798e0514054aac62f2a077252794ccb936572c2017ed5cd1ec98ff4576acecee
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_64000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1a02a98200eaa06575a1502e5da9c2523750268d07cb712a8bf14305e0c20c6
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_72000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72cdcc0410e7f0e9eadb7a993d874da8207eb8df86a67afca7bc34a37523c3c7
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_8000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:276c86afdb00e16ac411fb2392aaa4c1ae96a0723072f21bec023394f86f4415
+size 227724444

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/iter_80000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ffe6453b32f3af928bbda5aed4bb23c23a6368d6e82f5e8ab5bb3fb7f185bf4
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_logits/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ffe6453b32f3af928bbda5aed4bb23c23a6368d6e82f5e8ab5bb3fb7f185bf4
+size 227725084

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/20230304_103602.log

@@ -0,0 +1,1151 @@
+2023-03-04 10:36:02,337 - mmseg - INFO - Multi-processing start method is `None`
+2023-03-04 10:36:02,353 - mmseg - INFO - OpenCV num_threads is `128
+2023-03-04 10:36:02,353 - mmseg - INFO - OMP num threads is 1
+2023-03-04 10:36:02,407 - mmseg - INFO - Environment info:
+------------------------------------------------------------
+sys.platform: linux
+Python: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]
+CUDA available: True
+GPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB
+CUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch
+NVCC: Cuda compilation tools, release 11.6, V11.6.124
+GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)
+PyTorch: 1.13.1
+PyTorch compiling details: PyTorch built with:
+  - GCC 9.3
+  - C++ Version: 201402
+  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
+  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
+  - OpenMP 201511 (a.k.a. OpenMP 4.5)
+  - LAPACK is enabled (usually provided by MKL)
+  - NNPACK is enabled
+  - CPU capability usage: AVX2
+  - CUDA Runtime 11.6
+  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
+  - CuDNN 8.3.2  (built against CUDA 11.5)
+  - Magma 2.6.1
+  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 
+
+TorchVision: 0.14.1
+OpenCV: 4.7.0
+MMCV: 1.7.1
+MMCV Compiler: GCC 9.3
+MMCV CUDA Compiler: 11.6
+MMSegmentation: 0.30.0+d4f0cb3
+------------------------------------------------------------
+
+2023-03-04 10:36:02,407 - mmseg - INFO - Distributed training: True
+2023-03-04 10:36:03,072 - mmseg - INFO - Config:
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepMask',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=272,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask'
+gpu_ids = range(0, 8)
+auto_resume = True
+
+2023-03-04 10:36:07,359 - mmseg - INFO - Set random seed to 1470787464, deterministic: False
+2023-03-04 10:36:07,710 - mmseg - INFO - Parameters in backbone freezed!
+2023-03-04 10:36:07,712 - mmseg - INFO - Trainable parameters in SegformerHeadUnetFCHeadSingleStep: ['unet.init_conv.weight', 'unet.init_conv.bias', 'unet.time_mlp.1.weight', 'unet.time_mlp.1.bias', 'unet.time_mlp.3.weight', 'unet.time_mlp.3.bias', 'unet.downs.0.0.mlp.1.weight', 'unet.downs.0.0.mlp.1.bias', 'unet.downs.0.0.block1.proj.weight', 'unet.downs.0.0.block1.proj.bias', 'unet.downs.0.0.block1.norm.weight', 'unet.downs.0.0.block1.norm.bias', 'unet.downs.0.0.block2.proj.weight', 'unet.downs.0.0.block2.proj.bias', 'unet.downs.0.0.block2.norm.weight', 'unet.downs.0.0.block2.norm.bias', 'unet.downs.0.1.mlp.1.weight', 'unet.downs.0.1.mlp.1.bias', 'unet.downs.0.1.block1.proj.weight', 'unet.downs.0.1.block1.proj.bias', 'unet.downs.0.1.block1.norm.weight', 'unet.downs.0.1.block1.norm.bias', 'unet.downs.0.1.block2.proj.weight', 'unet.downs.0.1.block2.proj.bias', 'unet.downs.0.1.block2.norm.weight', 'unet.downs.0.1.block2.norm.bias', 'unet.downs.0.2.fn.fn.to_qkv.weight', 'unet.downs.0.2.fn.fn.to_out.0.weight', 'unet.downs.0.2.fn.fn.to_out.0.bias', 'unet.downs.0.2.fn.fn.to_out.1.g', 'unet.downs.0.2.fn.norm.g', 'unet.downs.0.3.weight', 'unet.downs.0.3.bias', 'unet.downs.1.0.mlp.1.weight', 'unet.downs.1.0.mlp.1.bias', 'unet.downs.1.0.block1.proj.weight', 'unet.downs.1.0.block1.proj.bias', 'unet.downs.1.0.block1.norm.weight', 'unet.downs.1.0.block1.norm.bias', 'unet.downs.1.0.block2.proj.weight', 'unet.downs.1.0.block2.proj.bias', 'unet.downs.1.0.block2.norm.weight', 'unet.downs.1.0.block2.norm.bias', 'unet.downs.1.1.mlp.1.weight', 'unet.downs.1.1.mlp.1.bias', 'unet.downs.1.1.block1.proj.weight', 'unet.downs.1.1.block1.proj.bias', 'unet.downs.1.1.block1.norm.weight', 'unet.downs.1.1.block1.norm.bias', 'unet.downs.1.1.block2.proj.weight', 'unet.downs.1.1.block2.proj.bias', 'unet.downs.1.1.block2.norm.weight', 'unet.downs.1.1.block2.norm.bias', 'unet.downs.1.2.fn.fn.to_qkv.weight', 'unet.downs.1.2.fn.fn.to_out.0.weight', 'unet.downs.1.2.fn.fn.to_out.0.bias', 'unet.downs.1.2.fn.fn.to_out.1.g', 'unet.downs.1.2.fn.norm.g', 'unet.downs.1.3.weight', 'unet.downs.1.3.bias', 'unet.downs.2.0.mlp.1.weight', 'unet.downs.2.0.mlp.1.bias', 'unet.downs.2.0.block1.proj.weight', 'unet.downs.2.0.block1.proj.bias', 'unet.downs.2.0.block1.norm.weight', 'unet.downs.2.0.block1.norm.bias', 'unet.downs.2.0.block2.proj.weight', 'unet.downs.2.0.block2.proj.bias', 'unet.downs.2.0.block2.norm.weight', 'unet.downs.2.0.block2.norm.bias', 'unet.downs.2.1.mlp.1.weight', 'unet.downs.2.1.mlp.1.bias', 'unet.downs.2.1.block1.proj.weight', 'unet.downs.2.1.block1.proj.bias', 'unet.downs.2.1.block1.norm.weight', 'unet.downs.2.1.block1.norm.bias', 'unet.downs.2.1.block2.proj.weight', 'unet.downs.2.1.block2.proj.bias', 'unet.downs.2.1.block2.norm.weight', 'unet.downs.2.1.block2.norm.bias', 'unet.downs.2.2.fn.fn.to_qkv.weight', 'unet.downs.2.2.fn.fn.to_out.0.weight', 'unet.downs.2.2.fn.fn.to_out.0.bias', 'unet.downs.2.2.fn.fn.to_out.1.g', 'unet.downs.2.2.fn.norm.g', 'unet.downs.2.3.weight', 'unet.downs.2.3.bias', 'unet.ups.0.0.mlp.1.weight', 'unet.ups.0.0.mlp.1.bias', 'unet.ups.0.0.block1.proj.weight', 'unet.ups.0.0.block1.proj.bias', 'unet.ups.0.0.block1.norm.weight', 'unet.ups.0.0.block1.norm.bias', 'unet.ups.0.0.block2.proj.weight', 'unet.ups.0.0.block2.proj.bias', 'unet.ups.0.0.block2.norm.weight', 'unet.ups.0.0.block2.norm.bias', 'unet.ups.0.0.res_conv.weight', 'unet.ups.0.0.res_conv.bias', 'unet.ups.0.1.mlp.1.weight', 'unet.ups.0.1.mlp.1.bias', 'unet.ups.0.1.block1.proj.weight', 'unet.ups.0.1.block1.proj.bias', 'unet.ups.0.1.block1.norm.weight', 'unet.ups.0.1.block1.norm.bias', 'unet.ups.0.1.block2.proj.weight', 'unet.ups.0.1.block2.proj.bias', 'unet.ups.0.1.block2.norm.weight', 'unet.ups.0.1.block2.norm.bias', 'unet.ups.0.1.res_conv.weight', 'unet.ups.0.1.res_conv.bias', 'unet.ups.0.2.fn.fn.to_qkv.weight', 'unet.ups.0.2.fn.fn.to_out.0.weight', 'unet.ups.0.2.fn.fn.to_out.0.bias', 'unet.ups.0.2.fn.fn.to_out.1.g', 'unet.ups.0.2.fn.norm.g', 'unet.ups.0.3.1.weight', 'unet.ups.0.3.1.bias', 'unet.ups.1.0.mlp.1.weight', 'unet.ups.1.0.mlp.1.bias', 'unet.ups.1.0.block1.proj.weight', 'unet.ups.1.0.block1.proj.bias', 'unet.ups.1.0.block1.norm.weight', 'unet.ups.1.0.block1.norm.bias', 'unet.ups.1.0.block2.proj.weight', 'unet.ups.1.0.block2.proj.bias', 'unet.ups.1.0.block2.norm.weight', 'unet.ups.1.0.block2.norm.bias', 'unet.ups.1.0.res_conv.weight', 'unet.ups.1.0.res_conv.bias', 'unet.ups.1.1.mlp.1.weight', 'unet.ups.1.1.mlp.1.bias', 'unet.ups.1.1.block1.proj.weight', 'unet.ups.1.1.block1.proj.bias', 'unet.ups.1.1.block1.norm.weight', 'unet.ups.1.1.block1.norm.bias', 'unet.ups.1.1.block2.proj.weight', 'unet.ups.1.1.block2.proj.bias', 'unet.ups.1.1.block2.norm.weight', 'unet.ups.1.1.block2.norm.bias', 'unet.ups.1.1.res_conv.weight', 'unet.ups.1.1.res_conv.bias', 'unet.ups.1.2.fn.fn.to_qkv.weight', 'unet.ups.1.2.fn.fn.to_out.0.weight', 'unet.ups.1.2.fn.fn.to_out.0.bias', 'unet.ups.1.2.fn.fn.to_out.1.g', 'unet.ups.1.2.fn.norm.g', 'unet.ups.1.3.1.weight', 'unet.ups.1.3.1.bias', 'unet.ups.2.0.mlp.1.weight', 'unet.ups.2.0.mlp.1.bias', 'unet.ups.2.0.block1.proj.weight', 'unet.ups.2.0.block1.proj.bias', 'unet.ups.2.0.block1.norm.weight', 'unet.ups.2.0.block1.norm.bias', 'unet.ups.2.0.block2.proj.weight', 'unet.ups.2.0.block2.proj.bias', 'unet.ups.2.0.block2.norm.weight', 'unet.ups.2.0.block2.norm.bias', 'unet.ups.2.0.res_conv.weight', 'unet.ups.2.0.res_conv.bias', 'unet.ups.2.1.mlp.1.weight', 'unet.ups.2.1.mlp.1.bias', 'unet.ups.2.1.block1.proj.weight', 'unet.ups.2.1.block1.proj.bias', 'unet.ups.2.1.block1.norm.weight', 'unet.ups.2.1.block1.norm.bias', 'unet.ups.2.1.block2.proj.weight', 'unet.ups.2.1.block2.proj.bias', 'unet.ups.2.1.block2.norm.weight', 'unet.ups.2.1.block2.norm.bias', 'unet.ups.2.1.res_conv.weight', 'unet.ups.2.1.res_conv.bias', 'unet.ups.2.2.fn.fn.to_qkv.weight', 'unet.ups.2.2.fn.fn.to_out.0.weight', 'unet.ups.2.2.fn.fn.to_out.0.bias', 'unet.ups.2.2.fn.fn.to_out.1.g', 'unet.ups.2.2.fn.norm.g', 'unet.ups.2.3.weight', 'unet.ups.2.3.bias', 'unet.mid_block1.mlp.1.weight', 'unet.mid_block1.mlp.1.bias', 'unet.mid_block1.block1.proj.weight', 'unet.mid_block1.block1.proj.bias', 'unet.mid_block1.block1.norm.weight', 'unet.mid_block1.block1.norm.bias', 'unet.mid_block1.block2.proj.weight', 'unet.mid_block1.block2.proj.bias', 'unet.mid_block1.block2.norm.weight', 'unet.mid_block1.block2.norm.bias', 'unet.mid_attn.fn.fn.to_qkv.weight', 'unet.mid_attn.fn.fn.to_out.weight', 'unet.mid_attn.fn.fn.to_out.bias', 'unet.mid_attn.fn.norm.g', 'unet.mid_block2.mlp.1.weight', 'unet.mid_block2.mlp.1.bias', 'unet.mid_block2.block1.proj.weight', 'unet.mid_block2.block1.proj.bias', 'unet.mid_block2.block1.norm.weight', 'unet.mid_block2.block1.norm.bias', 'unet.mid_block2.block2.proj.weight', 'unet.mid_block2.block2.proj.bias', 'unet.mid_block2.block2.norm.weight', 'unet.mid_block2.block2.norm.bias', 'unet.final_res_block.mlp.1.weight', 'unet.final_res_block.mlp.1.bias', 'unet.final_res_block.block1.proj.weight', 'unet.final_res_block.block1.proj.bias', 'unet.final_res_block.block1.norm.weight', 'unet.final_res_block.block1.norm.bias', 'unet.final_res_block.block2.proj.weight', 'unet.final_res_block.block2.proj.bias', 'unet.final_res_block.block2.norm.weight', 'unet.final_res_block.block2.norm.bias', 'unet.final_res_block.res_conv.weight', 'unet.final_res_block.res_conv.bias', 'unet.final_conv.weight', 'unet.final_conv.bias', 'conv_seg_new.weight', 'conv_seg_new.bias']
+2023-03-04 10:36:07,712 - mmseg - INFO - Parameters in decode_head freezed!
+2023-03-04 10:36:07,736 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 10:36:07,975 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: decode_head.conv_seg.weight, decode_head.conv_seg.bias, decode_head.convs.0.conv.weight, decode_head.convs.0.bn.weight, decode_head.convs.0.bn.bias, decode_head.convs.0.bn.running_mean, decode_head.convs.0.bn.running_var, decode_head.convs.0.bn.num_batches_tracked, decode_head.convs.1.conv.weight, decode_head.convs.1.bn.weight, decode_head.convs.1.bn.bias, decode_head.convs.1.bn.running_mean, decode_head.convs.1.bn.running_var, decode_head.convs.1.bn.num_batches_tracked, decode_head.convs.2.conv.weight, decode_head.convs.2.bn.weight, decode_head.convs.2.bn.bias, decode_head.convs.2.bn.running_mean, decode_head.convs.2.bn.running_var, decode_head.convs.2.bn.num_batches_tracked, decode_head.convs.3.conv.weight, decode_head.convs.3.bn.weight, decode_head.convs.3.bn.bias, decode_head.convs.3.bn.running_mean, decode_head.convs.3.bn.running_var, decode_head.convs.3.bn.num_batches_tracked, decode_head.fusion_conv.conv.weight, decode_head.fusion_conv.bn.weight, decode_head.fusion_conv.bn.bias, decode_head.fusion_conv.bn.running_mean, decode_head.fusion_conv.bn.running_var, decode_head.fusion_conv.bn.num_batches_tracked
+
+2023-03-04 10:36:07,990 - mmseg - INFO - load checkpoint from local path: pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth
+2023-03-04 10:36:08,180 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: backbone.layers.0.0.projection.weight, backbone.layers.0.0.projection.bias, backbone.layers.0.0.norm.weight, backbone.layers.0.0.norm.bias, backbone.layers.0.1.0.norm1.weight, backbone.layers.0.1.0.norm1.bias, backbone.layers.0.1.0.attn.attn.in_proj_weight, backbone.layers.0.1.0.attn.attn.in_proj_bias, backbone.layers.0.1.0.attn.attn.out_proj.weight, backbone.layers.0.1.0.attn.attn.out_proj.bias, backbone.layers.0.1.0.attn.sr.weight, backbone.layers.0.1.0.attn.sr.bias, backbone.layers.0.1.0.attn.norm.weight, backbone.layers.0.1.0.attn.norm.bias, backbone.layers.0.1.0.norm2.weight, backbone.layers.0.1.0.norm2.bias, backbone.layers.0.1.0.ffn.layers.0.weight, backbone.layers.0.1.0.ffn.layers.0.bias, backbone.layers.0.1.0.ffn.layers.1.weight, backbone.layers.0.1.0.ffn.layers.1.bias, backbone.layers.0.1.0.ffn.layers.4.weight, backbone.layers.0.1.0.ffn.layers.4.bias, backbone.layers.0.1.1.norm1.weight, backbone.layers.0.1.1.norm1.bias, backbone.layers.0.1.1.attn.attn.in_proj_weight, backbone.layers.0.1.1.attn.attn.in_proj_bias, backbone.layers.0.1.1.attn.attn.out_proj.weight, backbone.layers.0.1.1.attn.attn.out_proj.bias, backbone.layers.0.1.1.attn.sr.weight, backbone.layers.0.1.1.attn.sr.bias, backbone.layers.0.1.1.attn.norm.weight, backbone.layers.0.1.1.attn.norm.bias, backbone.layers.0.1.1.norm2.weight, backbone.layers.0.1.1.norm2.bias, backbone.layers.0.1.1.ffn.layers.0.weight, backbone.layers.0.1.1.ffn.layers.0.bias, backbone.layers.0.1.1.ffn.layers.1.weight, backbone.layers.0.1.1.ffn.layers.1.bias, backbone.layers.0.1.1.ffn.layers.4.weight, backbone.layers.0.1.1.ffn.layers.4.bias, backbone.layers.0.1.2.norm1.weight, backbone.layers.0.1.2.norm1.bias, backbone.layers.0.1.2.attn.attn.in_proj_weight, backbone.layers.0.1.2.attn.attn.in_proj_bias, backbone.layers.0.1.2.attn.attn.out_proj.weight, backbone.layers.0.1.2.attn.attn.out_proj.bias, backbone.layers.0.1.2.attn.sr.weight, backbone.layers.0.1.2.attn.sr.bias, backbone.layers.0.1.2.attn.norm.weight, backbone.layers.0.1.2.attn.norm.bias, backbone.layers.0.1.2.norm2.weight, backbone.layers.0.1.2.norm2.bias, backbone.layers.0.1.2.ffn.layers.0.weight, backbone.layers.0.1.2.ffn.layers.0.bias, backbone.layers.0.1.2.ffn.layers.1.weight, backbone.layers.0.1.2.ffn.layers.1.bias, backbone.layers.0.1.2.ffn.layers.4.weight, backbone.layers.0.1.2.ffn.layers.4.bias, backbone.layers.0.2.weight, backbone.layers.0.2.bias, backbone.layers.1.0.projection.weight, backbone.layers.1.0.projection.bias, backbone.layers.1.0.norm.weight, backbone.layers.1.0.norm.bias, backbone.layers.1.1.0.norm1.weight, backbone.layers.1.1.0.norm1.bias, backbone.layers.1.1.0.attn.attn.in_proj_weight, backbone.layers.1.1.0.attn.attn.in_proj_bias, backbone.layers.1.1.0.attn.attn.out_proj.weight, backbone.layers.1.1.0.attn.attn.out_proj.bias, backbone.layers.1.1.0.attn.sr.weight, backbone.layers.1.1.0.attn.sr.bias, backbone.layers.1.1.0.attn.norm.weight, backbone.layers.1.1.0.attn.norm.bias, backbone.layers.1.1.0.norm2.weight, backbone.layers.1.1.0.norm2.bias, backbone.layers.1.1.0.ffn.layers.0.weight, backbone.layers.1.1.0.ffn.layers.0.bias, backbone.layers.1.1.0.ffn.layers.1.weight, backbone.layers.1.1.0.ffn.layers.1.bias, backbone.layers.1.1.0.ffn.layers.4.weight, backbone.layers.1.1.0.ffn.layers.4.bias, backbone.layers.1.1.1.norm1.weight, backbone.layers.1.1.1.norm1.bias, backbone.layers.1.1.1.attn.attn.in_proj_weight, backbone.layers.1.1.1.attn.attn.in_proj_bias, backbone.layers.1.1.1.attn.attn.out_proj.weight, backbone.layers.1.1.1.attn.attn.out_proj.bias, backbone.layers.1.1.1.attn.sr.weight, backbone.layers.1.1.1.attn.sr.bias, backbone.layers.1.1.1.attn.norm.weight, backbone.layers.1.1.1.attn.norm.bias, backbone.layers.1.1.1.norm2.weight, backbone.layers.1.1.1.norm2.bias, backbone.layers.1.1.1.ffn.layers.0.weight, backbone.layers.1.1.1.ffn.layers.0.bias, backbone.layers.1.1.1.ffn.layers.1.weight, backbone.layers.1.1.1.ffn.layers.1.bias, backbone.layers.1.1.1.ffn.layers.4.weight, backbone.layers.1.1.1.ffn.layers.4.bias, backbone.layers.1.1.2.norm1.weight, backbone.layers.1.1.2.norm1.bias, backbone.layers.1.1.2.attn.attn.in_proj_weight, backbone.layers.1.1.2.attn.attn.in_proj_bias, backbone.layers.1.1.2.attn.attn.out_proj.weight, backbone.layers.1.1.2.attn.attn.out_proj.bias, backbone.layers.1.1.2.attn.sr.weight, backbone.layers.1.1.2.attn.sr.bias, backbone.layers.1.1.2.attn.norm.weight, backbone.layers.1.1.2.attn.norm.bias, backbone.layers.1.1.2.norm2.weight, backbone.layers.1.1.2.norm2.bias, backbone.layers.1.1.2.ffn.layers.0.weight, backbone.layers.1.1.2.ffn.layers.0.bias, backbone.layers.1.1.2.ffn.layers.1.weight, backbone.layers.1.1.2.ffn.layers.1.bias, backbone.layers.1.1.2.ffn.layers.4.weight, backbone.layers.1.1.2.ffn.layers.4.bias, backbone.layers.1.1.3.norm1.weight, backbone.layers.1.1.3.norm1.bias, backbone.layers.1.1.3.attn.attn.in_proj_weight, backbone.layers.1.1.3.attn.attn.in_proj_bias, backbone.layers.1.1.3.attn.attn.out_proj.weight, backbone.layers.1.1.3.attn.attn.out_proj.bias, backbone.layers.1.1.3.attn.sr.weight, backbone.layers.1.1.3.attn.sr.bias, backbone.layers.1.1.3.attn.norm.weight, backbone.layers.1.1.3.attn.norm.bias, backbone.layers.1.1.3.norm2.weight, backbone.layers.1.1.3.norm2.bias, backbone.layers.1.1.3.ffn.layers.0.weight, backbone.layers.1.1.3.ffn.layers.0.bias, backbone.layers.1.1.3.ffn.layers.1.weight, backbone.layers.1.1.3.ffn.layers.1.bias, backbone.layers.1.1.3.ffn.layers.4.weight, backbone.layers.1.1.3.ffn.layers.4.bias, backbone.layers.1.2.weight, backbone.layers.1.2.bias, backbone.layers.2.0.projection.weight, backbone.layers.2.0.projection.bias, backbone.layers.2.0.norm.weight, backbone.layers.2.0.norm.bias, backbone.layers.2.1.0.norm1.weight, backbone.layers.2.1.0.norm1.bias, backbone.layers.2.1.0.attn.attn.in_proj_weight, backbone.layers.2.1.0.attn.attn.in_proj_bias, backbone.layers.2.1.0.attn.attn.out_proj.weight, backbone.layers.2.1.0.attn.attn.out_proj.bias, backbone.layers.2.1.0.attn.sr.weight, backbone.layers.2.1.0.attn.sr.bias, backbone.layers.2.1.0.attn.norm.weight, backbone.layers.2.1.0.attn.norm.bias, backbone.layers.2.1.0.norm2.weight, backbone.layers.2.1.0.norm2.bias, backbone.layers.2.1.0.ffn.layers.0.weight, backbone.layers.2.1.0.ffn.layers.0.bias, backbone.layers.2.1.0.ffn.layers.1.weight, backbone.layers.2.1.0.ffn.layers.1.bias, backbone.layers.2.1.0.ffn.layers.4.weight, backbone.layers.2.1.0.ffn.layers.4.bias, backbone.layers.2.1.1.norm1.weight, backbone.layers.2.1.1.norm1.bias, backbone.layers.2.1.1.attn.attn.in_proj_weight, backbone.layers.2.1.1.attn.attn.in_proj_bias, backbone.layers.2.1.1.attn.attn.out_proj.weight, backbone.layers.2.1.1.attn.attn.out_proj.bias, backbone.layers.2.1.1.attn.sr.weight, backbone.layers.2.1.1.attn.sr.bias, backbone.layers.2.1.1.attn.norm.weight, backbone.layers.2.1.1.attn.norm.bias, backbone.layers.2.1.1.norm2.weight, backbone.layers.2.1.1.norm2.bias, backbone.layers.2.1.1.ffn.layers.0.weight, backbone.layers.2.1.1.ffn.layers.0.bias, backbone.layers.2.1.1.ffn.layers.1.weight, backbone.layers.2.1.1.ffn.layers.1.bias, backbone.layers.2.1.1.ffn.layers.4.weight, backbone.layers.2.1.1.ffn.layers.4.bias, backbone.layers.2.1.2.norm1.weight, backbone.layers.2.1.2.norm1.bias, backbone.layers.2.1.2.attn.attn.in_proj_weight, backbone.layers.2.1.2.attn.attn.in_proj_bias, backbone.layers.2.1.2.attn.attn.out_proj.weight, backbone.layers.2.1.2.attn.attn.out_proj.bias, backbone.layers.2.1.2.attn.sr.weight, backbone.layers.2.1.2.attn.sr.bias, backbone.layers.2.1.2.attn.norm.weight, backbone.layers.2.1.2.attn.norm.bias, backbone.layers.2.1.2.norm2.weight, backbone.layers.2.1.2.norm2.bias, backbone.layers.2.1.2.ffn.layers.0.weight, backbone.layers.2.1.2.ffn.layers.0.bias, backbone.layers.2.1.2.ffn.layers.1.weight, backbone.layers.2.1.2.ffn.layers.1.bias, backbone.layers.2.1.2.ffn.layers.4.weight, backbone.layers.2.1.2.ffn.layers.4.bias, backbone.layers.2.1.3.norm1.weight, backbone.layers.2.1.3.norm1.bias, backbone.layers.2.1.3.attn.attn.in_proj_weight, backbone.layers.2.1.3.attn.attn.in_proj_bias, backbone.layers.2.1.3.attn.attn.out_proj.weight, backbone.layers.2.1.3.attn.attn.out_proj.bias, backbone.layers.2.1.3.attn.sr.weight, backbone.layers.2.1.3.attn.sr.bias, backbone.layers.2.1.3.attn.norm.weight, backbone.layers.2.1.3.attn.norm.bias, backbone.layers.2.1.3.norm2.weight, backbone.layers.2.1.3.norm2.bias, backbone.layers.2.1.3.ffn.layers.0.weight, backbone.layers.2.1.3.ffn.layers.0.bias, backbone.layers.2.1.3.ffn.layers.1.weight, backbone.layers.2.1.3.ffn.layers.1.bias, backbone.layers.2.1.3.ffn.layers.4.weight, backbone.layers.2.1.3.ffn.layers.4.bias, backbone.layers.2.1.4.norm1.weight, backbone.layers.2.1.4.norm1.bias, backbone.layers.2.1.4.attn.attn.in_proj_weight, backbone.layers.2.1.4.attn.attn.in_proj_bias, backbone.layers.2.1.4.attn.attn.out_proj.weight, backbone.layers.2.1.4.attn.attn.out_proj.bias, backbone.layers.2.1.4.attn.sr.weight, backbone.layers.2.1.4.attn.sr.bias, backbone.layers.2.1.4.attn.norm.weight, backbone.layers.2.1.4.attn.norm.bias, backbone.layers.2.1.4.norm2.weight, backbone.layers.2.1.4.norm2.bias, backbone.layers.2.1.4.ffn.layers.0.weight, backbone.layers.2.1.4.ffn.layers.0.bias, backbone.layers.2.1.4.ffn.layers.1.weight, backbone.layers.2.1.4.ffn.layers.1.bias, backbone.layers.2.1.4.ffn.layers.4.weight, backbone.layers.2.1.4.ffn.layers.4.bias, backbone.layers.2.1.5.norm1.weight, backbone.layers.2.1.5.norm1.bias, backbone.layers.2.1.5.attn.attn.in_proj_weight, backbone.layers.2.1.5.attn.attn.in_proj_bias, backbone.layers.2.1.5.attn.attn.out_proj.weight, backbone.layers.2.1.5.attn.attn.out_proj.bias, backbone.layers.2.1.5.attn.sr.weight, backbone.layers.2.1.5.attn.sr.bias, backbone.layers.2.1.5.attn.norm.weight, backbone.layers.2.1.5.attn.norm.bias, backbone.layers.2.1.5.norm2.weight, backbone.layers.2.1.5.norm2.bias, backbone.layers.2.1.5.ffn.layers.0.weight, backbone.layers.2.1.5.ffn.layers.0.bias, backbone.layers.2.1.5.ffn.layers.1.weight, backbone.layers.2.1.5.ffn.layers.1.bias, backbone.layers.2.1.5.ffn.layers.4.weight, backbone.layers.2.1.5.ffn.layers.4.bias, backbone.layers.2.2.weight, backbone.layers.2.2.bias, backbone.layers.3.0.projection.weight, backbone.layers.3.0.projection.bias, backbone.layers.3.0.norm.weight, backbone.layers.3.0.norm.bias, backbone.layers.3.1.0.norm1.weight, backbone.layers.3.1.0.norm1.bias, backbone.layers.3.1.0.attn.attn.in_proj_weight, backbone.layers.3.1.0.attn.attn.in_proj_bias, backbone.layers.3.1.0.attn.attn.out_proj.weight, backbone.layers.3.1.0.attn.attn.out_proj.bias, backbone.layers.3.1.0.norm2.weight, backbone.layers.3.1.0.norm2.bias, backbone.layers.3.1.0.ffn.layers.0.weight, backbone.layers.3.1.0.ffn.layers.0.bias, backbone.layers.3.1.0.ffn.layers.1.weight, backbone.layers.3.1.0.ffn.layers.1.bias, backbone.layers.3.1.0.ffn.layers.4.weight, backbone.layers.3.1.0.ffn.layers.4.bias, backbone.layers.3.1.1.norm1.weight, backbone.layers.3.1.1.norm1.bias, backbone.layers.3.1.1.attn.attn.in_proj_weight, backbone.layers.3.1.1.attn.attn.in_proj_bias, backbone.layers.3.1.1.attn.attn.out_proj.weight, backbone.layers.3.1.1.attn.attn.out_proj.bias, backbone.layers.3.1.1.norm2.weight, backbone.layers.3.1.1.norm2.bias, backbone.layers.3.1.1.ffn.layers.0.weight, backbone.layers.3.1.1.ffn.layers.0.bias, backbone.layers.3.1.1.ffn.layers.1.weight, backbone.layers.3.1.1.ffn.layers.1.bias, backbone.layers.3.1.1.ffn.layers.4.weight, backbone.layers.3.1.1.ffn.layers.4.bias, backbone.layers.3.1.2.norm1.weight, backbone.layers.3.1.2.norm1.bias, backbone.layers.3.1.2.attn.attn.in_proj_weight, backbone.layers.3.1.2.attn.attn.in_proj_bias, backbone.layers.3.1.2.attn.attn.out_proj.weight, backbone.layers.3.1.2.attn.attn.out_proj.bias, backbone.layers.3.1.2.norm2.weight, backbone.layers.3.1.2.norm2.bias, backbone.layers.3.1.2.ffn.layers.0.weight, backbone.layers.3.1.2.ffn.layers.0.bias, backbone.layers.3.1.2.ffn.layers.1.weight, backbone.layers.3.1.2.ffn.layers.1.bias, backbone.layers.3.1.2.ffn.layers.4.weight, backbone.layers.3.1.2.ffn.layers.4.bias, backbone.layers.3.2.weight, backbone.layers.3.2.bias
+
+missing keys in source state_dict: unet.init_conv.weight, unet.init_conv.bias, unet.time_mlp.1.weight, unet.time_mlp.1.bias, unet.time_mlp.3.weight, unet.time_mlp.3.bias, unet.downs.0.0.mlp.1.weight, unet.downs.0.0.mlp.1.bias, unet.downs.0.0.block1.proj.weight, unet.downs.0.0.block1.proj.bias, unet.downs.0.0.block1.norm.weight, unet.downs.0.0.block1.norm.bias, unet.downs.0.0.block2.proj.weight, unet.downs.0.0.block2.proj.bias, unet.downs.0.0.block2.norm.weight, unet.downs.0.0.block2.norm.bias, unet.downs.0.1.mlp.1.weight, unet.downs.0.1.mlp.1.bias, unet.downs.0.1.block1.proj.weight, unet.downs.0.1.block1.proj.bias, unet.downs.0.1.block1.norm.weight, unet.downs.0.1.block1.norm.bias, unet.downs.0.1.block2.proj.weight, unet.downs.0.1.block2.proj.bias, unet.downs.0.1.block2.norm.weight, unet.downs.0.1.block2.norm.bias, unet.downs.0.2.fn.fn.to_qkv.weight, unet.downs.0.2.fn.fn.to_out.0.weight, unet.downs.0.2.fn.fn.to_out.0.bias, unet.downs.0.2.fn.fn.to_out.1.g, unet.downs.0.2.fn.norm.g, unet.downs.0.3.weight, unet.downs.0.3.bias, unet.downs.1.0.mlp.1.weight, unet.downs.1.0.mlp.1.bias, unet.downs.1.0.block1.proj.weight, unet.downs.1.0.block1.proj.bias, unet.downs.1.0.block1.norm.weight, unet.downs.1.0.block1.norm.bias, unet.downs.1.0.block2.proj.weight, unet.downs.1.0.block2.proj.bias, unet.downs.1.0.block2.norm.weight, unet.downs.1.0.block2.norm.bias, unet.downs.1.1.mlp.1.weight, unet.downs.1.1.mlp.1.bias, unet.downs.1.1.block1.proj.weight, unet.downs.1.1.block1.proj.bias, unet.downs.1.1.block1.norm.weight, unet.downs.1.1.block1.norm.bias, unet.downs.1.1.block2.proj.weight, unet.downs.1.1.block2.proj.bias, unet.downs.1.1.block2.norm.weight, unet.downs.1.1.block2.norm.bias, unet.downs.1.2.fn.fn.to_qkv.weight, unet.downs.1.2.fn.fn.to_out.0.weight, unet.downs.1.2.fn.fn.to_out.0.bias, unet.downs.1.2.fn.fn.to_out.1.g, unet.downs.1.2.fn.norm.g, unet.downs.1.3.weight, unet.downs.1.3.bias, unet.downs.2.0.mlp.1.weight, unet.downs.2.0.mlp.1.bias, unet.downs.2.0.block1.proj.weight, unet.downs.2.0.block1.proj.bias, unet.downs.2.0.block1.norm.weight, unet.downs.2.0.block1.norm.bias, unet.downs.2.0.block2.proj.weight, unet.downs.2.0.block2.proj.bias, unet.downs.2.0.block2.norm.weight, unet.downs.2.0.block2.norm.bias, unet.downs.2.1.mlp.1.weight, unet.downs.2.1.mlp.1.bias, unet.downs.2.1.block1.proj.weight, unet.downs.2.1.block1.proj.bias, unet.downs.2.1.block1.norm.weight, unet.downs.2.1.block1.norm.bias, unet.downs.2.1.block2.proj.weight, unet.downs.2.1.block2.proj.bias, unet.downs.2.1.block2.norm.weight, unet.downs.2.1.block2.norm.bias, unet.downs.2.2.fn.fn.to_qkv.weight, unet.downs.2.2.fn.fn.to_out.0.weight, unet.downs.2.2.fn.fn.to_out.0.bias, unet.downs.2.2.fn.fn.to_out.1.g, unet.downs.2.2.fn.norm.g, unet.downs.2.3.weight, unet.downs.2.3.bias, unet.ups.0.0.mlp.1.weight, unet.ups.0.0.mlp.1.bias, unet.ups.0.0.block1.proj.weight, unet.ups.0.0.block1.proj.bias, unet.ups.0.0.block1.norm.weight, unet.ups.0.0.block1.norm.bias, unet.ups.0.0.block2.proj.weight, unet.ups.0.0.block2.proj.bias, unet.ups.0.0.block2.norm.weight, unet.ups.0.0.block2.norm.bias, unet.ups.0.0.res_conv.weight, unet.ups.0.0.res_conv.bias, unet.ups.0.1.mlp.1.weight, unet.ups.0.1.mlp.1.bias, unet.ups.0.1.block1.proj.weight, unet.ups.0.1.block1.proj.bias, unet.ups.0.1.block1.norm.weight, unet.ups.0.1.block1.norm.bias, unet.ups.0.1.block2.proj.weight, unet.ups.0.1.block2.proj.bias, unet.ups.0.1.block2.norm.weight, unet.ups.0.1.block2.norm.bias, unet.ups.0.1.res_conv.weight, unet.ups.0.1.res_conv.bias, unet.ups.0.2.fn.fn.to_qkv.weight, unet.ups.0.2.fn.fn.to_out.0.weight, unet.ups.0.2.fn.fn.to_out.0.bias, unet.ups.0.2.fn.fn.to_out.1.g, unet.ups.0.2.fn.norm.g, unet.ups.0.3.1.weight, unet.ups.0.3.1.bias, unet.ups.1.0.mlp.1.weight, unet.ups.1.0.mlp.1.bias, unet.ups.1.0.block1.proj.weight, unet.ups.1.0.block1.proj.bias, unet.ups.1.0.block1.norm.weight, unet.ups.1.0.block1.norm.bias, unet.ups.1.0.block2.proj.weight, unet.ups.1.0.block2.proj.bias, unet.ups.1.0.block2.norm.weight, unet.ups.1.0.block2.norm.bias, unet.ups.1.0.res_conv.weight, unet.ups.1.0.res_conv.bias, unet.ups.1.1.mlp.1.weight, unet.ups.1.1.mlp.1.bias, unet.ups.1.1.block1.proj.weight, unet.ups.1.1.block1.proj.bias, unet.ups.1.1.block1.norm.weight, unet.ups.1.1.block1.norm.bias, unet.ups.1.1.block2.proj.weight, unet.ups.1.1.block2.proj.bias, unet.ups.1.1.block2.norm.weight, unet.ups.1.1.block2.norm.bias, unet.ups.1.1.res_conv.weight, unet.ups.1.1.res_conv.bias, unet.ups.1.2.fn.fn.to_qkv.weight, unet.ups.1.2.fn.fn.to_out.0.weight, unet.ups.1.2.fn.fn.to_out.0.bias, unet.ups.1.2.fn.fn.to_out.1.g, unet.ups.1.2.fn.norm.g, unet.ups.1.3.1.weight, unet.ups.1.3.1.bias, unet.ups.2.0.mlp.1.weight, unet.ups.2.0.mlp.1.bias, unet.ups.2.0.block1.proj.weight, unet.ups.2.0.block1.proj.bias, unet.ups.2.0.block1.norm.weight, unet.ups.2.0.block1.norm.bias, unet.ups.2.0.block2.proj.weight, unet.ups.2.0.block2.proj.bias, unet.ups.2.0.block2.norm.weight, unet.ups.2.0.block2.norm.bias, unet.ups.2.0.res_conv.weight, unet.ups.2.0.res_conv.bias, unet.ups.2.1.mlp.1.weight, unet.ups.2.1.mlp.1.bias, unet.ups.2.1.block1.proj.weight, unet.ups.2.1.block1.proj.bias, unet.ups.2.1.block1.norm.weight, unet.ups.2.1.block1.norm.bias, unet.ups.2.1.block2.proj.weight, unet.ups.2.1.block2.proj.bias, unet.ups.2.1.block2.norm.weight, unet.ups.2.1.block2.norm.bias, unet.ups.2.1.res_conv.weight, unet.ups.2.1.res_conv.bias, unet.ups.2.2.fn.fn.to_qkv.weight, unet.ups.2.2.fn.fn.to_out.0.weight, unet.ups.2.2.fn.fn.to_out.0.bias, unet.ups.2.2.fn.fn.to_out.1.g, unet.ups.2.2.fn.norm.g, unet.ups.2.3.weight, unet.ups.2.3.bias, unet.mid_block1.mlp.1.weight, unet.mid_block1.mlp.1.bias, unet.mid_block1.block1.proj.weight, unet.mid_block1.block1.proj.bias, unet.mid_block1.block1.norm.weight, unet.mid_block1.block1.norm.bias, unet.mid_block1.block2.proj.weight, unet.mid_block1.block2.proj.bias, unet.mid_block1.block2.norm.weight, unet.mid_block1.block2.norm.bias, unet.mid_attn.fn.fn.to_qkv.weight, unet.mid_attn.fn.fn.to_out.weight, unet.mid_attn.fn.fn.to_out.bias, unet.mid_attn.fn.norm.g, unet.mid_block2.mlp.1.weight, unet.mid_block2.mlp.1.bias, unet.mid_block2.block1.proj.weight, unet.mid_block2.block1.proj.bias, unet.mid_block2.block1.norm.weight, unet.mid_block2.block1.norm.bias, unet.mid_block2.block2.proj.weight, unet.mid_block2.block2.proj.bias, unet.mid_block2.block2.norm.weight, unet.mid_block2.block2.norm.bias, unet.final_res_block.mlp.1.weight, unet.final_res_block.mlp.1.bias, unet.final_res_block.block1.proj.weight, unet.final_res_block.block1.proj.bias, unet.final_res_block.block1.norm.weight, unet.final_res_block.block1.norm.bias, unet.final_res_block.block2.proj.weight, unet.final_res_block.block2.proj.bias, unet.final_res_block.block2.norm.weight, unet.final_res_block.block2.norm.bias, unet.final_res_block.res_conv.weight, unet.final_res_block.res_conv.bias, unet.final_conv.weight, unet.final_conv.bias, conv_seg_new.weight, conv_seg_new.bias, embed.weight
+
+2023-03-04 10:36:08,206 - mmseg - INFO - EncoderDecoderFreeze(
+  (backbone): MixVisionTransformerCustomInitWeights(
+    (layers): ModuleList(
+      (0): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(3, 64, kernel_size=(7, 7), stride=(4, 4), padding=(3, 3))
+          (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+      )
+      (1): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+      )
+      (2): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(128, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (4): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (5): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+      )
+      (3): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(320, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+      )
+    )
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+  (decode_head): SegformerHeadUnetFCHeadSingleStepMask(
+    input_transform=multiple_select, ignore_index=0, align_corners=False
+    (loss_decode): CrossEntropyLoss(avg_non_ignore=False)
+    (conv_seg): None
+    (dropout): Dropout2d(p=0.1, inplace=False)
+    (convs): ModuleList(
+      (0): ConvModule(
+        (conv): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (1): ConvModule(
+        (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (2): ConvModule(
+        (conv): Conv2d(320, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (3): ConvModule(
+        (conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+    )
+    (fusion_conv): ConvModule(
+      (conv): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+      (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+      (activate): ReLU(inplace=True)
+    )
+    (unet): Unet(
+      (init_conv): Conv2d(272, 128, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3))
+      (time_mlp): Sequential(
+        (0): SinusoidalPosEmb()
+        (1): Linear(in_features=128, out_features=512, bias=True)
+        (2): GELU(approximate='none')
+        (3): Linear(in_features=512, out_features=512, bias=True)
+      )
+      (downs): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (ups): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (mid_block1): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (mid_attn): Residual(
+        (fn): PreNorm(
+          (fn): Attention(
+            (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+            (to_out): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (norm): LayerNorm()
+        )
+      )
+      (mid_block2): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (final_res_block): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+      )
+      (final_conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1))
+    )
+    (conv_seg_new): Conv2d(256, 151, kernel_size=(1, 1), stride=(1, 1))
+    (embed): Embedding(152, 16)
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'}
+)
+2023-03-04 10:36:09,087 - mmseg - INFO - Loaded 20210 images
+2023-03-04 10:36:10,053 - mmseg - INFO - Loaded 2000 images
+2023-03-04 10:36:10,056 - mmseg - INFO - Start running, host: laizeqiang@SH-IDC1-10-140-37-113, work_dir: /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask
+2023-03-04 10:36:10,056 - mmseg - INFO - Hooks will be executed in the following order:
+before_run:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_epoch:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_iter:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+ -------------------- 
+after_train_iter:
+(ABOVE_NORMAL) OptimizerHook                      
+(NORMAL      ) CheckpointHook                     
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_train_epoch:
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_epoch:
+(LOW         ) IterTimerHook                      
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_epoch:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_run:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+2023-03-04 10:36:10,056 - mmseg - INFO - workflow: [('train', 1)], max: 80000 iters
+2023-03-04 10:36:10,056 - mmseg - INFO - Checkpoints will be saved to /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask by HardDiskBackend.
+2023-03-04 10:36:47,431 - mmseg - INFO - Iter [50/80000]	lr: 7.350e-06, eta: 6:21:07, time: 0.286, data_time: 0.014, memory: 19783, decode.loss_ce: 3.8243, decode.acc_seg: 14.2062, loss: 3.8243
+2023-03-04 10:36:55,947 - mmseg - INFO - Iter [100/80000]	lr: 1.485e-05, eta: 5:03:50, time: 0.170, data_time: 0.007, memory: 19783, decode.loss_ce: 2.9748, decode.acc_seg: 42.5617, loss: 2.9748
+2023-03-04 10:37:04,369 - mmseg - INFO - Iter [150/80000]	lr: 2.235e-05, eta: 4:37:05, time: 0.168, data_time: 0.008, memory: 19783, decode.loss_ce: 2.2347, decode.acc_seg: 49.3174, loss: 2.2347
+2023-03-04 10:37:12,840 - mmseg - INFO - Iter [200/80000]	lr: 2.985e-05, eta: 4:24:00, time: 0.169, data_time: 0.007, memory: 19783, decode.loss_ce: 1.6971, decode.acc_seg: 60.8327, loss: 1.6971
+2023-03-04 10:37:21,248 - mmseg - INFO - Iter [250/80000]	lr: 3.735e-05, eta: 4:15:49, time: 0.168, data_time: 0.007, memory: 19783, decode.loss_ce: 1.3266, decode.acc_seg: 69.1492, loss: 1.3266
+2023-03-04 10:37:29,768 - mmseg - INFO - Iter [300/80000]	lr: 4.485e-05, eta: 4:10:46, time: 0.170, data_time: 0.007, memory: 19783, decode.loss_ce: 1.1094, decode.acc_seg: 74.3359, loss: 1.1094
+2023-03-04 10:37:38,291 - mmseg - INFO - Iter [350/80000]	lr: 5.235e-05, eta: 4:07:08, time: 0.170, data_time: 0.007, memory: 19783, decode.loss_ce: 0.8799, decode.acc_seg: 78.0994, loss: 0.8799
+2023-03-04 10:37:46,865 - mmseg - INFO - Iter [400/80000]	lr: 5.985e-05, eta: 4:04:32, time: 0.171, data_time: 0.007, memory: 19783, decode.loss_ce: 0.7519, decode.acc_seg: 80.2631, loss: 0.7519
+2023-03-04 10:37:55,418 - mmseg - INFO - Iter [450/80000]	lr: 6.735e-05, eta: 4:02:26, time: 0.171, data_time: 0.007, memory: 19783, decode.loss_ce: 0.6707, decode.acc_seg: 81.7875, loss: 0.6707
+2023-03-04 10:38:03,650 - mmseg - INFO - Iter [500/80000]	lr: 7.485e-05, eta: 3:59:52, time: 0.165, data_time: 0.007, memory: 19783, decode.loss_ce: 0.5952, decode.acc_seg: 82.6875, loss: 0.5952
+2023-03-04 10:38:12,334 - mmseg - INFO - Iter [550/80000]	lr: 8.235e-05, eta: 3:58:49, time: 0.174, data_time: 0.008, memory: 19783, decode.loss_ce: 0.5100, decode.acc_seg: 84.7401, loss: 0.5100
+2023-03-04 10:38:20,396 - mmseg - INFO - Iter [600/80000]	lr: 8.985e-05, eta: 3:56:34, time: 0.161, data_time: 0.008, memory: 19783, decode.loss_ce: 0.4445, decode.acc_seg: 85.9801, loss: 0.4445
+2023-03-04 10:38:31,315 - mmseg - INFO - Iter [650/80000]	lr: 9.735e-05, eta: 4:00:27, time: 0.218, data_time: 0.054, memory: 19783, decode.loss_ce: 0.4351, decode.acc_seg: 86.1025, loss: 0.4351
+2023-03-04 10:38:39,745 - mmseg - INFO - Iter [700/80000]	lr: 1.049e-04, eta: 3:59:03, time: 0.169, data_time: 0.007, memory: 19783, decode.loss_ce: 0.4012, decode.acc_seg: 86.6512, loss: 0.4012

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/20230304_103602.log.json

@@ -0,0 +1,15 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+d4f0cb3", "seed": 1470787464, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask.py", "mmseg_version": "0.30.0+d4f0cb3", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepMask',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=272,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 1470787464\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}
+{"mode": "train", "epoch": 1, "iter": 50, "lr": 1e-05, "memory": 19783, "data_time": 0.01408, "decode.loss_ce": 3.82431, "decode.acc_seg": 14.20625, "loss": 3.82431, "time": 0.28603}
+{"mode": "train", "epoch": 1, "iter": 100, "lr": 1e-05, "memory": 19783, "data_time": 0.0072, "decode.loss_ce": 2.9748, "decode.acc_seg": 42.56172, "loss": 2.9748, "time": 0.17031}
+{"mode": "train", "epoch": 1, "iter": 150, "lr": 2e-05, "memory": 19783, "data_time": 0.00767, "decode.loss_ce": 2.23466, "decode.acc_seg": 49.31739, "loss": 2.23466, "time": 0.16828}
+{"mode": "train", "epoch": 1, "iter": 200, "lr": 3e-05, "memory": 19783, "data_time": 0.00714, "decode.loss_ce": 1.69707, "decode.acc_seg": 60.8327, "loss": 1.69707, "time": 0.16941}
+{"mode": "train", "epoch": 1, "iter": 250, "lr": 4e-05, "memory": 19783, "data_time": 0.00685, "decode.loss_ce": 1.32658, "decode.acc_seg": 69.1492, "loss": 1.32658, "time": 0.16832}
+{"mode": "train", "epoch": 1, "iter": 300, "lr": 4e-05, "memory": 19783, "data_time": 0.00661, "decode.loss_ce": 1.10944, "decode.acc_seg": 74.3359, "loss": 1.10944, "time": 0.17039}
+{"mode": "train", "epoch": 1, "iter": 350, "lr": 5e-05, "memory": 19783, "data_time": 0.00731, "decode.loss_ce": 0.87994, "decode.acc_seg": 78.09939, "loss": 0.87994, "time": 0.17046}
+{"mode": "train", "epoch": 1, "iter": 400, "lr": 6e-05, "memory": 19783, "data_time": 0.00667, "decode.loss_ce": 0.75186, "decode.acc_seg": 80.26307, "loss": 0.75186, "time": 0.17146}
+{"mode": "train", "epoch": 1, "iter": 450, "lr": 7e-05, "memory": 19783, "data_time": 0.00738, "decode.loss_ce": 0.67067, "decode.acc_seg": 81.78752, "loss": 0.67067, "time": 0.17104}
+{"mode": "train", "epoch": 1, "iter": 500, "lr": 7e-05, "memory": 19783, "data_time": 0.00748, "decode.loss_ce": 0.59517, "decode.acc_seg": 82.68755, "loss": 0.59517, "time": 0.16462}
+{"mode": "train", "epoch": 1, "iter": 550, "lr": 8e-05, "memory": 19783, "data_time": 0.00755, "decode.loss_ce": 0.51003, "decode.acc_seg": 84.74014, "loss": 0.51003, "time": 0.17368}
+{"mode": "train", "epoch": 1, "iter": 600, "lr": 9e-05, "memory": 19783, "data_time": 0.00752, "decode.loss_ce": 0.44453, "decode.acc_seg": 85.98014, "loss": 0.44453, "time": 0.16123}
+{"mode": "train", "epoch": 2, "iter": 650, "lr": 0.0001, "memory": 19783, "data_time": 0.05353, "decode.loss_ce": 0.43514, "decode.acc_seg": 86.10254, "loss": 0.43514, "time": 0.21836}
+{"mode": "train", "epoch": 2, "iter": 700, "lr": 0.0001, "memory": 19783, "data_time": 0.00717, "decode.loss_ce": 0.40124, "decode.acc_seg": 86.65119, "loss": 0.40124, "time": 0.16862}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/20230304_103934.log.json

@@ -0,0 +1,161 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+d4f0cb3", "seed": 1648012630, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask.py", "mmseg_version": "0.30.0+d4f0cb3", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\nmodel = dict(\n    type='EncoderDecoderFreeze',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadSingleStepMask',\n        pretrained=\n        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=272,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=10000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=80000)\ncheckpoint_config = dict(by_epoch=False, interval=8000)\nevaluation = dict(\n    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 1648012630\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}
+{"mode": "train", "epoch": 1, "iter": 50, "lr": 1e-05, "memory": 19783, "data_time": 0.01464, "decode.loss_ce": 3.78586, "decode.acc_seg": 13.44899, "loss": 3.78586, "time": 0.28897}
+{"mode": "train", "epoch": 1, "iter": 100, "lr": 1e-05, "memory": 19783, "data_time": 0.00682, "decode.loss_ce": 2.91953, "decode.acc_seg": 44.1399, "loss": 2.91953, "time": 0.17252}
+{"mode": "train", "epoch": 1, "iter": 150, "lr": 2e-05, "memory": 19783, "data_time": 0.00701, "decode.loss_ce": 2.10409, "decode.acc_seg": 53.58975, "loss": 2.10409, "time": 0.17548}
+{"mode": "train", "epoch": 1, "iter": 200, "lr": 3e-05, "memory": 19783, "data_time": 0.0073, "decode.loss_ce": 1.6526, "decode.acc_seg": 62.47836, "loss": 1.6526, "time": 0.16673}
+{"mode": "train", "epoch": 1, "iter": 250, "lr": 4e-05, "memory": 19783, "data_time": 0.00784, "decode.loss_ce": 1.32199, "decode.acc_seg": 68.99333, "loss": 1.32199, "time": 0.18392}
+{"mode": "train", "epoch": 1, "iter": 300, "lr": 4e-05, "memory": 19783, "data_time": 0.00685, "decode.loss_ce": 1.07112, "decode.acc_seg": 74.76253, "loss": 1.07112, "time": 0.16585}
+{"mode": "train", "epoch": 1, "iter": 350, "lr": 5e-05, "memory": 19783, "data_time": 0.00832, "decode.loss_ce": 0.87855, "decode.acc_seg": 77.97355, "loss": 0.87855, "time": 0.17731}
+{"mode": "train", "epoch": 1, "iter": 400, "lr": 6e-05, "memory": 19783, "data_time": 0.00796, "decode.loss_ce": 0.75946, "decode.acc_seg": 80.1142, "loss": 0.75946, "time": 0.16399}
+{"mode": "train", "epoch": 1, "iter": 450, "lr": 7e-05, "memory": 19783, "data_time": 0.00752, "decode.loss_ce": 0.70385, "decode.acc_seg": 80.5049, "loss": 0.70385, "time": 0.16518}
+{"mode": "train", "epoch": 1, "iter": 500, "lr": 7e-05, "memory": 19783, "data_time": 0.00758, "decode.loss_ce": 0.56582, "decode.acc_seg": 84.02578, "loss": 0.56582, "time": 0.16899}
+{"mode": "train", "epoch": 1, "iter": 550, "lr": 8e-05, "memory": 19783, "data_time": 0.00745, "decode.loss_ce": 0.51044, "decode.acc_seg": 84.65132, "loss": 0.51044, "time": 0.17316}
+{"mode": "train", "epoch": 1, "iter": 600, "lr": 9e-05, "memory": 19783, "data_time": 0.00692, "decode.loss_ce": 0.46692, "decode.acc_seg": 85.42044, "loss": 0.46692, "time": 0.17047}
+{"mode": "train", "epoch": 2, "iter": 650, "lr": 0.0001, "memory": 19783, "data_time": 0.05563, "decode.loss_ce": 0.42618, "decode.acc_seg": 86.16397, "loss": 0.42618, "time": 0.22341}
+{"mode": "train", "epoch": 2, "iter": 700, "lr": 0.0001, "memory": 19783, "data_time": 0.00702, "decode.loss_ce": 0.39092, "decode.acc_seg": 86.64982, "loss": 0.39092, "time": 0.16783}
+{"mode": "train", "epoch": 2, "iter": 750, "lr": 0.00011, "memory": 19783, "data_time": 0.00746, "decode.loss_ce": 0.35796, "decode.acc_seg": 87.63995, "loss": 0.35796, "time": 0.16982}
+{"mode": "train", "epoch": 2, "iter": 800, "lr": 0.00012, "memory": 19783, "data_time": 0.00735, "decode.loss_ce": 0.37266, "decode.acc_seg": 87.08973, "loss": 0.37266, "time": 0.16822}
+{"mode": "train", "epoch": 2, "iter": 850, "lr": 0.00013, "memory": 19783, "data_time": 0.00717, "decode.loss_ce": 0.3519, "decode.acc_seg": 87.47414, "loss": 0.3519, "time": 0.17154}
+{"mode": "train", "epoch": 2, "iter": 900, "lr": 0.00013, "memory": 19783, "data_time": 0.00814, "decode.loss_ce": 0.33697, "decode.acc_seg": 87.83325, "loss": 0.33697, "time": 0.17741}
+{"mode": "train", "epoch": 2, "iter": 950, "lr": 0.00014, "memory": 19783, "data_time": 0.007, "decode.loss_ce": 0.33214, "decode.acc_seg": 87.90779, "loss": 0.33214, "time": 0.17481}
+{"mode": "train", "epoch": 2, "iter": 1000, "lr": 0.00015, "memory": 19783, "data_time": 0.00707, "decode.loss_ce": 0.32533, "decode.acc_seg": 87.80602, "loss": 0.32533, "time": 0.17805}
+{"mode": "train", "epoch": 2, "iter": 1050, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.31727, "decode.acc_seg": 88.31456, "loss": 0.31727, "time": 0.16702}
+{"mode": "train", "epoch": 2, "iter": 1100, "lr": 0.00015, "memory": 19783, "data_time": 0.00683, "decode.loss_ce": 0.31942, "decode.acc_seg": 88.159, "loss": 0.31942, "time": 0.1682}
+{"mode": "train", "epoch": 2, "iter": 1150, "lr": 0.00015, "memory": 19783, "data_time": 0.0074, "decode.loss_ce": 0.31511, "decode.acc_seg": 88.10372, "loss": 0.31511, "time": 0.16781}
+{"mode": "train", "epoch": 2, "iter": 1200, "lr": 0.00015, "memory": 19783, "data_time": 0.00735, "decode.loss_ce": 0.29375, "decode.acc_seg": 89.09847, "loss": 0.29375, "time": 0.16453}
+{"mode": "train", "epoch": 2, "iter": 1250, "lr": 0.00015, "memory": 19783, "data_time": 0.00714, "decode.loss_ce": 0.29113, "decode.acc_seg": 88.97655, "loss": 0.29113, "time": 0.16846}
+{"mode": "train", "epoch": 3, "iter": 1300, "lr": 0.00015, "memory": 19783, "data_time": 0.05392, "decode.loss_ce": 0.28411, "decode.acc_seg": 89.29744, "loss": 0.28411, "time": 0.21186}
+{"mode": "train", "epoch": 3, "iter": 1350, "lr": 0.00015, "memory": 19783, "data_time": 0.0071, "decode.loss_ce": 0.28474, "decode.acc_seg": 88.92053, "loss": 0.28474, "time": 0.16269}
+{"mode": "train", "epoch": 3, "iter": 1400, "lr": 0.00015, "memory": 19783, "data_time": 0.00671, "decode.loss_ce": 0.28921, "decode.acc_seg": 89.11701, "loss": 0.28921, "time": 0.16276}
+{"mode": "train", "epoch": 3, "iter": 1450, "lr": 0.00015, "memory": 19783, "data_time": 0.00739, "decode.loss_ce": 0.28266, "decode.acc_seg": 89.08791, "loss": 0.28266, "time": 0.16348}
+{"mode": "train", "epoch": 3, "iter": 1500, "lr": 0.00015, "memory": 19783, "data_time": 0.00699, "decode.loss_ce": 0.28323, "decode.acc_seg": 88.94225, "loss": 0.28323, "time": 0.16677}
+{"mode": "train", "epoch": 3, "iter": 1550, "lr": 0.00015, "memory": 19783, "data_time": 0.0075, "decode.loss_ce": 0.29209, "decode.acc_seg": 88.851, "loss": 0.29209, "time": 0.16656}
+{"mode": "train", "epoch": 3, "iter": 1600, "lr": 0.00015, "memory": 19783, "data_time": 0.00727, "decode.loss_ce": 0.28026, "decode.acc_seg": 89.16342, "loss": 0.28026, "time": 0.16899}
+{"mode": "train", "epoch": 3, "iter": 1650, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.27397, "decode.acc_seg": 89.47841, "loss": 0.27397, "time": 0.17142}
+{"mode": "train", "epoch": 3, "iter": 1700, "lr": 0.00015, "memory": 19783, "data_time": 0.00748, "decode.loss_ce": 0.27943, "decode.acc_seg": 89.04867, "loss": 0.27943, "time": 0.17033}
+{"mode": "train", "epoch": 3, "iter": 1750, "lr": 0.00015, "memory": 19783, "data_time": 0.00739, "decode.loss_ce": 0.27289, "decode.acc_seg": 89.30963, "loss": 0.27289, "time": 0.16775}
+{"mode": "train", "epoch": 3, "iter": 1800, "lr": 0.00015, "memory": 19783, "data_time": 0.00709, "decode.loss_ce": 0.25431, "decode.acc_seg": 89.92175, "loss": 0.25431, "time": 0.17115}
+{"mode": "train", "epoch": 3, "iter": 1850, "lr": 0.00015, "memory": 19783, "data_time": 0.00764, "decode.loss_ce": 0.27098, "decode.acc_seg": 89.38316, "loss": 0.27098, "time": 0.16868}
+{"mode": "train", "epoch": 4, "iter": 1900, "lr": 0.00015, "memory": 19783, "data_time": 0.05451, "decode.loss_ce": 0.26261, "decode.acc_seg": 89.79397, "loss": 0.26261, "time": 0.2239}
+{"mode": "train", "epoch": 4, "iter": 1950, "lr": 0.00015, "memory": 19783, "data_time": 0.00677, "decode.loss_ce": 0.27538, "decode.acc_seg": 89.28754, "loss": 0.27538, "time": 0.17293}
+{"mode": "train", "epoch": 4, "iter": 2000, "lr": 0.00015, "memory": 19783, "data_time": 0.00683, "decode.loss_ce": 0.26254, "decode.acc_seg": 89.72531, "loss": 0.26254, "time": 0.16631}
+{"mode": "train", "epoch": 4, "iter": 2050, "lr": 0.00015, "memory": 19783, "data_time": 0.00672, "decode.loss_ce": 0.26892, "decode.acc_seg": 89.60717, "loss": 0.26892, "time": 0.17634}
+{"mode": "train", "epoch": 4, "iter": 2100, "lr": 0.00015, "memory": 19783, "data_time": 0.0071, "decode.loss_ce": 0.25522, "decode.acc_seg": 89.88685, "loss": 0.25522, "time": 0.16423}
+{"mode": "train", "epoch": 4, "iter": 2150, "lr": 0.00015, "memory": 19783, "data_time": 0.00737, "decode.loss_ce": 0.26096, "decode.acc_seg": 89.79101, "loss": 0.26096, "time": 0.16638}
+{"mode": "train", "epoch": 4, "iter": 2200, "lr": 0.00015, "memory": 19783, "data_time": 0.00718, "decode.loss_ce": 0.26273, "decode.acc_seg": 89.645, "loss": 0.26273, "time": 0.17165}
+{"mode": "train", "epoch": 4, "iter": 2250, "lr": 0.00015, "memory": 19783, "data_time": 0.00694, "decode.loss_ce": 0.26784, "decode.acc_seg": 89.34889, "loss": 0.26784, "time": 0.16663}
+{"mode": "train", "epoch": 4, "iter": 2300, "lr": 0.00015, "memory": 19783, "data_time": 0.00715, "decode.loss_ce": 0.26124, "decode.acc_seg": 89.65508, "loss": 0.26124, "time": 0.17081}
+{"mode": "train", "epoch": 4, "iter": 2350, "lr": 0.00015, "memory": 19783, "data_time": 0.00701, "decode.loss_ce": 0.26513, "decode.acc_seg": 89.44922, "loss": 0.26513, "time": 0.16596}
+{"mode": "train", "epoch": 4, "iter": 2400, "lr": 0.00015, "memory": 19783, "data_time": 0.00661, "decode.loss_ce": 0.25128, "decode.acc_seg": 89.97749, "loss": 0.25128, "time": 0.17423}
+{"mode": "train", "epoch": 4, "iter": 2450, "lr": 0.00015, "memory": 19783, "data_time": 0.00745, "decode.loss_ce": 0.26641, "decode.acc_seg": 89.34108, "loss": 0.26641, "time": 0.17297}
+{"mode": "train", "epoch": 4, "iter": 2500, "lr": 0.00015, "memory": 19783, "data_time": 0.0072, "decode.loss_ce": 0.2621, "decode.acc_seg": 89.7861, "loss": 0.2621, "time": 0.17021}
+{"mode": "train", "epoch": 5, "iter": 2550, "lr": 0.00015, "memory": 19783, "data_time": 0.0521, "decode.loss_ce": 0.24304, "decode.acc_seg": 90.10812, "loss": 0.24304, "time": 0.21542}
+{"mode": "train", "epoch": 5, "iter": 2600, "lr": 0.00015, "memory": 19783, "data_time": 0.00686, "decode.loss_ce": 0.25384, "decode.acc_seg": 90.02081, "loss": 0.25384, "time": 0.17489}
+{"mode": "train", "epoch": 5, "iter": 2650, "lr": 0.00015, "memory": 19783, "data_time": 0.00716, "decode.loss_ce": 0.26031, "decode.acc_seg": 89.77231, "loss": 0.26031, "time": 0.16684}
+{"mode": "train", "epoch": 5, "iter": 2700, "lr": 0.00015, "memory": 19783, "data_time": 0.0071, "decode.loss_ce": 0.2666, "decode.acc_seg": 89.22362, "loss": 0.2666, "time": 0.16719}
+{"mode": "train", "epoch": 5, "iter": 2750, "lr": 0.00015, "memory": 19783, "data_time": 0.00801, "decode.loss_ce": 0.26155, "decode.acc_seg": 89.54207, "loss": 0.26155, "time": 0.16738}
+{"mode": "train", "epoch": 5, "iter": 2800, "lr": 0.00015, "memory": 19783, "data_time": 0.00727, "decode.loss_ce": 0.2504, "decode.acc_seg": 89.91151, "loss": 0.2504, "time": 0.16287}
+{"mode": "train", "epoch": 5, "iter": 2850, "lr": 0.00015, "memory": 19783, "data_time": 0.00674, "decode.loss_ce": 0.25434, "decode.acc_seg": 89.87366, "loss": 0.25434, "time": 0.17361}
+{"mode": "train", "epoch": 5, "iter": 2900, "lr": 0.00015, "memory": 19783, "data_time": 0.00791, "decode.loss_ce": 0.2511, "decode.acc_seg": 89.8458, "loss": 0.2511, "time": 0.17571}
+{"mode": "train", "epoch": 5, "iter": 2950, "lr": 0.00015, "memory": 19783, "data_time": 0.00664, "decode.loss_ce": 0.25013, "decode.acc_seg": 90.12326, "loss": 0.25013, "time": 0.17105}
+{"mode": "train", "epoch": 5, "iter": 3000, "lr": 0.00015, "memory": 19783, "data_time": 0.00693, "decode.loss_ce": 0.25603, "decode.acc_seg": 89.82247, "loss": 0.25603, "time": 0.1639}
+{"mode": "train", "epoch": 5, "iter": 3050, "lr": 0.00015, "memory": 19783, "data_time": 0.00718, "decode.loss_ce": 0.25956, "decode.acc_seg": 89.82373, "loss": 0.25956, "time": 0.17946}
+{"mode": "train", "epoch": 5, "iter": 3100, "lr": 0.00015, "memory": 19783, "data_time": 0.00707, "decode.loss_ce": 0.25529, "decode.acc_seg": 89.77675, "loss": 0.25529, "time": 0.16852}
+{"mode": "train", "epoch": 5, "iter": 3150, "lr": 0.00015, "memory": 19783, "data_time": 0.00737, "decode.loss_ce": 0.25803, "decode.acc_seg": 89.64798, "loss": 0.25803, "time": 0.17423}
+{"mode": "train", "epoch": 6, "iter": 3200, "lr": 0.00015, "memory": 19783, "data_time": 0.05491, "decode.loss_ce": 0.24815, "decode.acc_seg": 90.10166, "loss": 0.24815, "time": 0.21712}
+{"mode": "train", "epoch": 6, "iter": 3250, "lr": 0.00015, "memory": 19783, "data_time": 0.00732, "decode.loss_ce": 0.24073, "decode.acc_seg": 90.3456, "loss": 0.24073, "time": 0.1644}
+{"mode": "train", "epoch": 6, "iter": 3300, "lr": 0.00015, "memory": 19783, "data_time": 0.00742, "decode.loss_ce": 0.24323, "decode.acc_seg": 90.39054, "loss": 0.24323, "time": 0.16306}
+{"mode": "train", "epoch": 6, "iter": 3350, "lr": 0.00015, "memory": 19783, "data_time": 0.00727, "decode.loss_ce": 0.24765, "decode.acc_seg": 90.05928, "loss": 0.24765, "time": 0.16683}
+{"mode": "train", "epoch": 6, "iter": 3400, "lr": 0.00015, "memory": 19783, "data_time": 0.00744, "decode.loss_ce": 0.24116, "decode.acc_seg": 90.39199, "loss": 0.24116, "time": 0.16243}
+{"mode": "train", "epoch": 6, "iter": 3450, "lr": 0.00015, "memory": 19783, "data_time": 0.00781, "decode.loss_ce": 0.24984, "decode.acc_seg": 89.77229, "loss": 0.24984, "time": 0.16597}
+{"mode": "train", "epoch": 6, "iter": 3500, "lr": 0.00015, "memory": 19783, "data_time": 0.00758, "decode.loss_ce": 0.2521, "decode.acc_seg": 89.79541, "loss": 0.2521, "time": 0.16949}
+{"mode": "train", "epoch": 6, "iter": 3550, "lr": 0.00015, "memory": 19783, "data_time": 0.00802, "decode.loss_ce": 0.26202, "decode.acc_seg": 89.40474, "loss": 0.26202, "time": 0.16849}
+{"mode": "train", "epoch": 6, "iter": 3600, "lr": 0.00015, "memory": 19783, "data_time": 0.00764, "decode.loss_ce": 0.24962, "decode.acc_seg": 89.7942, "loss": 0.24962, "time": 0.16373}
+{"mode": "train", "epoch": 6, "iter": 3650, "lr": 0.00015, "memory": 19783, "data_time": 0.00706, "decode.loss_ce": 0.23575, "decode.acc_seg": 90.55114, "loss": 0.23575, "time": 0.1724}
+{"mode": "train", "epoch": 6, "iter": 3700, "lr": 0.00015, "memory": 19783, "data_time": 0.00707, "decode.loss_ce": 0.25319, "decode.acc_seg": 89.95363, "loss": 0.25319, "time": 0.17091}
+{"mode": "train", "epoch": 6, "iter": 3750, "lr": 0.00015, "memory": 19783, "data_time": 0.00681, "decode.loss_ce": 0.25331, "decode.acc_seg": 89.74242, "loss": 0.25331, "time": 0.1678}
+{"mode": "train", "epoch": 7, "iter": 3800, "lr": 0.00015, "memory": 19783, "data_time": 0.05404, "decode.loss_ce": 0.24787, "decode.acc_seg": 89.94419, "loss": 0.24787, "time": 0.23071}
+{"mode": "train", "epoch": 7, "iter": 3850, "lr": 0.00015, "memory": 19783, "data_time": 0.00738, "decode.loss_ce": 0.23945, "decode.acc_seg": 90.18102, "loss": 0.23945, "time": 0.1685}
+{"mode": "train", "epoch": 7, "iter": 3900, "lr": 0.00015, "memory": 19783, "data_time": 0.0076, "decode.loss_ce": 0.23457, "decode.acc_seg": 90.46158, "loss": 0.23457, "time": 0.17543}
+{"mode": "train", "epoch": 7, "iter": 3950, "lr": 0.00015, "memory": 19783, "data_time": 0.0076, "decode.loss_ce": 0.24769, "decode.acc_seg": 90.01688, "loss": 0.24769, "time": 0.16782}
+{"mode": "train", "epoch": 7, "iter": 4000, "lr": 0.00015, "memory": 19783, "data_time": 0.00717, "decode.loss_ce": 0.24628, "decode.acc_seg": 90.21013, "loss": 0.24628, "time": 0.16726}
+{"mode": "train", "epoch": 7, "iter": 4050, "lr": 0.00015, "memory": 19783, "data_time": 0.00754, "decode.loss_ce": 0.2341, "decode.acc_seg": 90.579, "loss": 0.2341, "time": 0.17162}
+{"mode": "train", "epoch": 7, "iter": 4100, "lr": 0.00015, "memory": 19783, "data_time": 0.00746, "decode.loss_ce": 0.24573, "decode.acc_seg": 90.16542, "loss": 0.24573, "time": 0.16339}
+{"mode": "train", "epoch": 7, "iter": 4150, "lr": 0.00015, "memory": 19783, "data_time": 0.00731, "decode.loss_ce": 0.24285, "decode.acc_seg": 90.28304, "loss": 0.24285, "time": 0.16864}
+{"mode": "train", "epoch": 7, "iter": 4200, "lr": 0.00015, "memory": 19783, "data_time": 0.00756, "decode.loss_ce": 0.23958, "decode.acc_seg": 90.30665, "loss": 0.23958, "time": 0.16759}
+{"mode": "train", "epoch": 7, "iter": 4250, "lr": 0.00015, "memory": 19783, "data_time": 0.00689, "decode.loss_ce": 0.24788, "decode.acc_seg": 90.00322, "loss": 0.24788, "time": 0.16993}
+{"mode": "train", "epoch": 7, "iter": 4300, "lr": 0.00015, "memory": 19783, "data_time": 0.00734, "decode.loss_ce": 0.23125, "decode.acc_seg": 90.54874, "loss": 0.23125, "time": 0.16811}
+{"mode": "train", "epoch": 7, "iter": 4350, "lr": 0.00015, "memory": 19783, "data_time": 0.00728, "decode.loss_ce": 0.24042, "decode.acc_seg": 90.47029, "loss": 0.24042, "time": 0.16826}
+{"mode": "train", "epoch": 7, "iter": 4400, "lr": 0.00015, "memory": 19783, "data_time": 0.00747, "decode.loss_ce": 0.24551, "decode.acc_seg": 89.95885, "loss": 0.24551, "time": 0.17054}
+{"mode": "train", "epoch": 8, "iter": 4450, "lr": 0.00015, "memory": 19783, "data_time": 0.05442, "decode.loss_ce": 0.2265, "decode.acc_seg": 90.90395, "loss": 0.2265, "time": 0.21336}
+{"mode": "train", "epoch": 8, "iter": 4500, "lr": 0.00015, "memory": 19783, "data_time": 0.00733, "decode.loss_ce": 0.247, "decode.acc_seg": 90.04266, "loss": 0.247, "time": 0.16689}
+{"mode": "train", "epoch": 8, "iter": 4550, "lr": 0.00015, "memory": 19783, "data_time": 0.0076, "decode.loss_ce": 0.2394, "decode.acc_seg": 90.46837, "loss": 0.2394, "time": 0.17266}
+{"mode": "train", "epoch": 8, "iter": 4600, "lr": 0.00015, "memory": 19783, "data_time": 0.00739, "decode.loss_ce": 0.24674, "decode.acc_seg": 90.0508, "loss": 0.24674, "time": 0.16544}
+{"mode": "train", "epoch": 8, "iter": 4650, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.24889, "decode.acc_seg": 89.97996, "loss": 0.24889, "time": 0.17361}
+{"mode": "train", "epoch": 8, "iter": 4700, "lr": 0.00015, "memory": 19783, "data_time": 0.00723, "decode.loss_ce": 0.25066, "decode.acc_seg": 90.10819, "loss": 0.25066, "time": 0.17526}
+{"mode": "train", "epoch": 8, "iter": 4750, "lr": 0.00015, "memory": 19783, "data_time": 0.00725, "decode.loss_ce": 0.23078, "decode.acc_seg": 90.68572, "loss": 0.23078, "time": 0.16917}
+{"mode": "train", "epoch": 8, "iter": 4800, "lr": 0.00015, "memory": 19783, "data_time": 0.00754, "decode.loss_ce": 0.24164, "decode.acc_seg": 90.21678, "loss": 0.24164, "time": 0.16936}
+{"mode": "train", "epoch": 8, "iter": 4850, "lr": 0.00015, "memory": 19783, "data_time": 0.00749, "decode.loss_ce": 0.24123, "decode.acc_seg": 90.2115, "loss": 0.24123, "time": 0.17225}
+{"mode": "train", "epoch": 8, "iter": 4900, "lr": 0.00015, "memory": 19783, "data_time": 0.00787, "decode.loss_ce": 0.24872, "decode.acc_seg": 90.06069, "loss": 0.24872, "time": 0.16871}
+{"mode": "train", "epoch": 8, "iter": 4950, "lr": 0.00015, "memory": 19783, "data_time": 0.00731, "decode.loss_ce": 0.24157, "decode.acc_seg": 90.28286, "loss": 0.24157, "time": 0.1622}
+{"mode": "train", "epoch": 8, "iter": 5000, "lr": 0.00015, "memory": 19783, "data_time": 0.00723, "decode.loss_ce": 0.23888, "decode.acc_seg": 90.337, "loss": 0.23888, "time": 0.17729}
+{"mode": "train", "epoch": 9, "iter": 5050, "lr": 0.00015, "memory": 19783, "data_time": 0.05372, "decode.loss_ce": 0.23444, "decode.acc_seg": 90.65998, "loss": 0.23444, "time": 0.21327}
+{"mode": "train", "epoch": 9, "iter": 5100, "lr": 0.00015, "memory": 19783, "data_time": 0.00674, "decode.loss_ce": 0.24385, "decode.acc_seg": 90.19634, "loss": 0.24385, "time": 0.17532}
+{"mode": "train", "epoch": 9, "iter": 5150, "lr": 0.00015, "memory": 19783, "data_time": 0.00754, "decode.loss_ce": 0.24576, "decode.acc_seg": 90.23702, "loss": 0.24576, "time": 0.16991}
+{"mode": "train", "epoch": 9, "iter": 5200, "lr": 0.00015, "memory": 19783, "data_time": 0.007, "decode.loss_ce": 0.2333, "decode.acc_seg": 90.51888, "loss": 0.2333, "time": 0.16263}
+{"mode": "train", "epoch": 9, "iter": 5250, "lr": 0.00015, "memory": 19783, "data_time": 0.00714, "decode.loss_ce": 0.24466, "decode.acc_seg": 90.27321, "loss": 0.24466, "time": 0.16789}
+{"mode": "train", "epoch": 9, "iter": 5300, "lr": 0.00015, "memory": 19783, "data_time": 0.0078, "decode.loss_ce": 0.24114, "decode.acc_seg": 90.39097, "loss": 0.24114, "time": 0.16739}
+{"mode": "train", "epoch": 9, "iter": 5350, "lr": 0.00015, "memory": 19783, "data_time": 0.0073, "decode.loss_ce": 0.23671, "decode.acc_seg": 90.32327, "loss": 0.23671, "time": 0.1613}
+{"mode": "train", "epoch": 9, "iter": 5400, "lr": 0.00015, "memory": 19783, "data_time": 0.00732, "decode.loss_ce": 0.23675, "decode.acc_seg": 90.48409, "loss": 0.23675, "time": 0.16479}
+{"mode": "train", "epoch": 9, "iter": 5450, "lr": 0.00015, "memory": 19783, "data_time": 0.00771, "decode.loss_ce": 0.24392, "decode.acc_seg": 90.07503, "loss": 0.24392, "time": 0.16243}
+{"mode": "train", "epoch": 9, "iter": 5500, "lr": 0.00015, "memory": 19783, "data_time": 0.0071, "decode.loss_ce": 0.2508, "decode.acc_seg": 89.99546, "loss": 0.2508, "time": 0.17402}
+{"mode": "train", "epoch": 9, "iter": 5550, "lr": 0.00015, "memory": 19783, "data_time": 0.00696, "decode.loss_ce": 0.24176, "decode.acc_seg": 90.22726, "loss": 0.24176, "time": 0.17048}
+{"mode": "train", "epoch": 9, "iter": 5600, "lr": 0.00015, "memory": 19783, "data_time": 0.00667, "decode.loss_ce": 0.23641, "decode.acc_seg": 90.32053, "loss": 0.23641, "time": 0.17978}
+{"mode": "train", "epoch": 9, "iter": 5650, "lr": 0.00015, "memory": 19783, "data_time": 0.008, "decode.loss_ce": 0.22384, "decode.acc_seg": 90.89573, "loss": 0.22384, "time": 0.17304}
+{"mode": "train", "epoch": 10, "iter": 5700, "lr": 0.00015, "memory": 19783, "data_time": 0.05376, "decode.loss_ce": 0.23805, "decode.acc_seg": 90.35932, "loss": 0.23805, "time": 0.21666}
+{"mode": "train", "epoch": 10, "iter": 5750, "lr": 0.00015, "memory": 19783, "data_time": 0.00731, "decode.loss_ce": 0.23584, "decode.acc_seg": 90.47929, "loss": 0.23584, "time": 0.16638}
+{"mode": "train", "epoch": 10, "iter": 5800, "lr": 0.00015, "memory": 19783, "data_time": 0.00671, "decode.loss_ce": 0.25196, "decode.acc_seg": 89.81801, "loss": 0.25196, "time": 0.16849}
+{"mode": "train", "epoch": 10, "iter": 5850, "lr": 0.00015, "memory": 19783, "data_time": 0.00707, "decode.loss_ce": 0.23571, "decode.acc_seg": 90.36798, "loss": 0.23571, "time": 0.17485}
+{"mode": "train", "epoch": 10, "iter": 5900, "lr": 0.00015, "memory": 19783, "data_time": 0.00676, "decode.loss_ce": 0.2416, "decode.acc_seg": 90.27747, "loss": 0.2416, "time": 0.17023}
+{"mode": "train", "epoch": 10, "iter": 5950, "lr": 0.00015, "memory": 19783, "data_time": 0.00715, "decode.loss_ce": 0.23559, "decode.acc_seg": 90.43211, "loss": 0.23559, "time": 0.17164}
+{"mode": "train", "epoch": 10, "iter": 6000, "lr": 0.00015, "memory": 19783, "data_time": 0.00697, "decode.loss_ce": 0.2205, "decode.acc_seg": 91.05051, "loss": 0.2205, "time": 0.17556}
+{"mode": "train", "epoch": 10, "iter": 6050, "lr": 0.00015, "memory": 19783, "data_time": 0.00694, "decode.loss_ce": 0.23222, "decode.acc_seg": 90.57611, "loss": 0.23222, "time": 0.17327}
+{"mode": "train", "epoch": 10, "iter": 6100, "lr": 0.00015, "memory": 19783, "data_time": 0.00714, "decode.loss_ce": 0.24064, "decode.acc_seg": 90.31305, "loss": 0.24064, "time": 0.16579}
+{"mode": "train", "epoch": 10, "iter": 6150, "lr": 0.00015, "memory": 19783, "data_time": 0.00751, "decode.loss_ce": 0.23976, "decode.acc_seg": 90.38844, "loss": 0.23976, "time": 0.16872}
+{"mode": "train", "epoch": 10, "iter": 6200, "lr": 0.00015, "memory": 19783, "data_time": 0.00733, "decode.loss_ce": 0.23414, "decode.acc_seg": 90.44385, "loss": 0.23414, "time": 0.16422}
+{"mode": "train", "epoch": 10, "iter": 6250, "lr": 0.00015, "memory": 19783, "data_time": 0.00727, "decode.loss_ce": 0.24566, "decode.acc_seg": 90.27457, "loss": 0.24566, "time": 0.17081}
+{"mode": "train", "epoch": 10, "iter": 6300, "lr": 0.00015, "memory": 19783, "data_time": 0.00741, "decode.loss_ce": 0.22406, "decode.acc_seg": 90.73291, "loss": 0.22406, "time": 0.172}
+{"mode": "train", "epoch": 11, "iter": 6350, "lr": 0.00015, "memory": 19783, "data_time": 0.05658, "decode.loss_ce": 0.23578, "decode.acc_seg": 90.38232, "loss": 0.23578, "time": 0.21898}
+{"mode": "train", "epoch": 11, "iter": 6400, "lr": 0.00015, "memory": 19783, "data_time": 0.00723, "decode.loss_ce": 0.22355, "decode.acc_seg": 90.76867, "loss": 0.22355, "time": 0.17097}
+{"mode": "train", "epoch": 11, "iter": 6450, "lr": 0.00015, "memory": 19783, "data_time": 0.00727, "decode.loss_ce": 0.22216, "decode.acc_seg": 90.82901, "loss": 0.22216, "time": 0.17446}
+{"mode": "train", "epoch": 11, "iter": 6500, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.22971, "decode.acc_seg": 90.70459, "loss": 0.22971, "time": 0.17473}
+{"mode": "train", "epoch": 11, "iter": 6550, "lr": 0.00015, "memory": 19783, "data_time": 0.00783, "decode.loss_ce": 0.22432, "decode.acc_seg": 90.83311, "loss": 0.22432, "time": 0.1701}
+{"mode": "train", "epoch": 11, "iter": 6600, "lr": 0.00015, "memory": 19783, "data_time": 0.00738, "decode.loss_ce": 0.24216, "decode.acc_seg": 90.42721, "loss": 0.24216, "time": 0.16582}
+{"mode": "train", "epoch": 11, "iter": 6650, "lr": 0.00015, "memory": 19783, "data_time": 0.00664, "decode.loss_ce": 0.23376, "decode.acc_seg": 90.60508, "loss": 0.23376, "time": 0.17496}
+{"mode": "train", "epoch": 11, "iter": 6700, "lr": 0.00015, "memory": 19783, "data_time": 0.00714, "decode.loss_ce": 0.24306, "decode.acc_seg": 90.1269, "loss": 0.24306, "time": 0.16614}
+{"mode": "train", "epoch": 11, "iter": 6750, "lr": 0.00015, "memory": 19783, "data_time": 0.00718, "decode.loss_ce": 0.23822, "decode.acc_seg": 90.38081, "loss": 0.23822, "time": 0.16329}
+{"mode": "train", "epoch": 11, "iter": 6800, "lr": 0.00015, "memory": 19783, "data_time": 0.00724, "decode.loss_ce": 0.2379, "decode.acc_seg": 90.49255, "loss": 0.2379, "time": 0.16966}
+{"mode": "train", "epoch": 11, "iter": 6850, "lr": 0.00015, "memory": 19783, "data_time": 0.00708, "decode.loss_ce": 0.24489, "decode.acc_seg": 90.19225, "loss": 0.24489, "time": 0.16918}
+{"mode": "train", "epoch": 11, "iter": 6900, "lr": 0.00015, "memory": 19783, "data_time": 0.00737, "decode.loss_ce": 0.23591, "decode.acc_seg": 90.47066, "loss": 0.23591, "time": 0.16545}
+{"mode": "train", "epoch": 12, "iter": 6950, "lr": 0.00015, "memory": 19783, "data_time": 0.05516, "decode.loss_ce": 0.22491, "decode.acc_seg": 90.74824, "loss": 0.22491, "time": 0.22021}
+{"mode": "train", "epoch": 12, "iter": 7000, "lr": 0.00015, "memory": 19783, "data_time": 0.00698, "decode.loss_ce": 0.22866, "decode.acc_seg": 90.84347, "loss": 0.22866, "time": 0.16929}
+{"mode": "train", "epoch": 12, "iter": 7050, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.23445, "decode.acc_seg": 90.53955, "loss": 0.23445, "time": 0.16909}
+{"mode": "train", "epoch": 12, "iter": 7100, "lr": 0.00015, "memory": 19783, "data_time": 0.00688, "decode.loss_ce": 0.22867, "decode.acc_seg": 90.72875, "loss": 0.22867, "time": 0.17427}
+{"mode": "train", "epoch": 12, "iter": 7150, "lr": 0.00015, "memory": 19783, "data_time": 0.00774, "decode.loss_ce": 0.23175, "decode.acc_seg": 90.64853, "loss": 0.23175, "time": 0.16845}
+{"mode": "train", "epoch": 12, "iter": 7200, "lr": 0.00015, "memory": 19783, "data_time": 0.00753, "decode.loss_ce": 0.23831, "decode.acc_seg": 90.29184, "loss": 0.23831, "time": 0.17243}
+{"mode": "train", "epoch": 12, "iter": 7250, "lr": 0.00015, "memory": 19783, "data_time": 0.00724, "decode.loss_ce": 0.23129, "decode.acc_seg": 90.67923, "loss": 0.23129, "time": 0.16382}
+{"mode": "train", "epoch": 12, "iter": 7300, "lr": 0.00015, "memory": 19783, "data_time": 0.00728, "decode.loss_ce": 0.23042, "decode.acc_seg": 90.6756, "loss": 0.23042, "time": 0.17514}
+{"mode": "train", "epoch": 12, "iter": 7350, "lr": 0.00015, "memory": 19783, "data_time": 0.00762, "decode.loss_ce": 0.23096, "decode.acc_seg": 90.4861, "loss": 0.23096, "time": 0.17494}
+{"mode": "train", "epoch": 12, "iter": 7400, "lr": 0.00015, "memory": 19783, "data_time": 0.00722, "decode.loss_ce": 0.23648, "decode.acc_seg": 90.41459, "loss": 0.23648, "time": 0.16612}
+{"mode": "train", "epoch": 12, "iter": 7450, "lr": 0.00015, "memory": 19783, "data_time": 0.00698, "decode.loss_ce": 0.23389, "decode.acc_seg": 90.56373, "loss": 0.23389, "time": 0.16687}
+{"mode": "train", "epoch": 12, "iter": 7500, "lr": 0.00015, "memory": 19783, "data_time": 0.00707, "decode.loss_ce": 0.22833, "decode.acc_seg": 90.66262, "loss": 0.22833, "time": 0.16645}
+{"mode": "train", "epoch": 12, "iter": 7550, "lr": 0.00015, "memory": 19783, "data_time": 0.00708, "decode.loss_ce": 0.23725, "decode.acc_seg": 90.30798, "loss": 0.23725, "time": 0.16547}
+{"mode": "train", "epoch": 13, "iter": 7600, "lr": 0.00015, "memory": 19783, "data_time": 0.05558, "decode.loss_ce": 0.23152, "decode.acc_seg": 90.56859, "loss": 0.23152, "time": 0.22103}
+{"mode": "train", "epoch": 13, "iter": 7650, "lr": 0.00015, "memory": 19783, "data_time": 0.00695, "decode.loss_ce": 0.23743, "decode.acc_seg": 90.40197, "loss": 0.23743, "time": 0.17243}
+{"mode": "train", "epoch": 13, "iter": 7700, "lr": 0.00015, "memory": 19783, "data_time": 0.0071, "decode.loss_ce": 0.23772, "decode.acc_seg": 90.38538, "loss": 0.23772, "time": 0.16691}
+{"mode": "train", "epoch": 13, "iter": 7750, "lr": 0.00015, "memory": 19783, "data_time": 0.00726, "decode.loss_ce": 0.23096, "decode.acc_seg": 90.64128, "loss": 0.23096, "time": 0.16896}
+{"mode": "train", "epoch": 13, "iter": 7800, "lr": 0.00015, "memory": 19783, "data_time": 0.00713, "decode.loss_ce": 0.23442, "decode.acc_seg": 90.45397, "loss": 0.23442, "time": 0.16189}
+{"mode": "train", "epoch": 13, "iter": 7850, "lr": 0.00015, "memory": 19783, "data_time": 0.007, "decode.loss_ce": 0.23541, "decode.acc_seg": 90.56818, "loss": 0.23541, "time": 0.16867}
+{"mode": "train", "epoch": 13, "iter": 7900, "lr": 0.00015, "memory": 19783, "data_time": 0.00689, "decode.loss_ce": 0.23165, "decode.acc_seg": 90.62381, "loss": 0.23165, "time": 0.16922}
+{"mode": "train", "epoch": 13, "iter": 7950, "lr": 0.00015, "memory": 19783, "data_time": 0.0072, "decode.loss_ce": 0.22949, "decode.acc_seg": 90.78639, "loss": 0.22949, "time": 0.16628}
+{"mode": "train", "epoch": 13, "iter": 8000, "lr": 0.00015, "memory": 19783, "data_time": 0.00678, "decode.loss_ce": 0.22824, "decode.acc_seg": 90.75584, "loss": 0.22824, "time": 0.18786}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask.py

@@ -0,0 +1,184 @@
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth'
+model = dict(
+    type='EncoderDecoderFreeze',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadSingleStepMask',
+        pretrained=
+        'pretrained/segformer_mit-b2_512x512_160k_ade20k_20220620_114047-64e4feca.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=272,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=10000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=80000)
+checkpoint_config = dict(by_epoch=False, interval=8000)
+evaluation = dict(
+    interval=8000, metric='mIoU', pre_eval=True, save_best='mIoU')
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask'
+gpu_ids = range(0, 8)
+auto_resume = True

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/best_mIoU_iter_80000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:081c08833e6ff1558fecbbb2faf46fb5d8eac1ac84e8262ffdcc37b15a7a0a14
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_16000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d535ddd93d708f0cb46b2d4f94066a6e7b3e28a15c5ba3e149ea7faa8d23f91e
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_24000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9d00a51da87942936bd198abc14b563649c0b16cf50b05a213c851e5796d9f0f
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_32000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7275faca1dd9925c1cc880b13a61d9e85021f7312bdcd0b6aa56091e1aa0a5f8
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_40000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7df7c1d42c842cfab9edbd862afd1922dccf0335b32af1f83faa24483a1c8191
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_48000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da5506d445951321beeb2ed61005415e06a3ca9f286236fa597d067971954695
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_56000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8cd4ede6001fea5831fe892f2ba8b8e222ccf8019c592ed0763b443b86847f97
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_64000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae9768a3d8d2a7b90cb594ceca93ab5233c99ad513cc404b86ebb4c68801d807
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_72000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63d77ed6173ae933c1ba225b175550faeb7e09f1e1ae334dbce6437d88cec317
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_8000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:584332478ca3ec6ccb3370d1bdb91adc01484d87ffbec918c777050cda731c53
+size 235547678

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/iter_80000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca82e7718d63dce0a800ea751b899f4f38cf0d4788a4378c5dc07a616e783d41
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151_mask/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca82e7718d63dce0a800ea751b899f4f38cf0d4788a4378c5dc07a616e783d41
+size 235548318

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/20230305_231050.log

@@ -0,0 +1,1152 @@
+2023-03-05 23:10:50,099 - mmseg - INFO - Multi-processing start method is `None`
+2023-03-05 23:10:50,117 - mmseg - INFO - OpenCV num_threads is `128
+2023-03-05 23:10:50,117 - mmseg - INFO - OMP num threads is 1
+2023-03-05 23:10:50,169 - mmseg - INFO - Environment info:
+------------------------------------------------------------
+sys.platform: linux
+Python: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]
+CUDA available: True
+GPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB
+CUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch
+NVCC: Cuda compilation tools, release 11.6, V11.6.124
+GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)
+PyTorch: 1.13.1
+PyTorch compiling details: PyTorch built with:
+  - GCC 9.3
+  - C++ Version: 201402
+  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications
+  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)
+  - OpenMP 201511 (a.k.a. OpenMP 4.5)
+  - LAPACK is enabled (usually provided by MKL)
+  - NNPACK is enabled
+  - CPU capability usage: AVX2
+  - CUDA Runtime 11.6
+  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37
+  - CuDNN 8.3.2  (built against CUDA 11.5)
+  - Magma 2.6.1
+  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 
+
+TorchVision: 0.14.1
+OpenCV: 4.7.0
+MMCV: 1.7.1
+MMCV Compiler: GCC 9.3
+MMCV CUDA Compiler: 11.6
+MMSegmentation: 0.30.0+6db5ece
+------------------------------------------------------------
+
+2023-03-05 23:10:50,169 - mmseg - INFO - Distributed training: True
+2023-03-05 23:10:50,859 - mmseg - INFO - Config:
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'
+model = dict(
+    type='EncoderDecoderDiffusion',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadMultiStepCE',
+        pretrained=
+        'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=272,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        diffusion_timesteps=100,
+        collect_timesteps=[0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 99],
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.1)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=20000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=160000)
+checkpoint_config = dict(by_epoch=False, interval=16000, max_keep_ckpts=1)
+evaluation = dict(
+    interval=16000, metric='mIoU', pre_eval=True, save_best='mIoU')
+custom_hooks = [
+    dict(
+        type='ConstantMomentumEMAHook',
+        momentum=0.01,
+        interval=25,
+        eval_interval=16000,
+        auto_resume=True,
+        priority=49)
+]
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce'
+gpu_ids = range(0, 8)
+auto_resume = True
+
+2023-03-05 23:10:55,198 - mmseg - INFO - Set random seed to 1580901347, deterministic: False
+2023-03-05 23:10:55,464 - mmseg - INFO - Parameters in backbone freezed!
+2023-03-05 23:10:55,465 - mmseg - INFO - Trainable parameters in SegformerHeadUnetFCHeadMultiStep: ['unet.init_conv.weight', 'unet.init_conv.bias', 'unet.time_mlp.1.weight', 'unet.time_mlp.1.bias', 'unet.time_mlp.3.weight', 'unet.time_mlp.3.bias', 'unet.downs.0.0.mlp.1.weight', 'unet.downs.0.0.mlp.1.bias', 'unet.downs.0.0.block1.proj.weight', 'unet.downs.0.0.block1.proj.bias', 'unet.downs.0.0.block1.norm.weight', 'unet.downs.0.0.block1.norm.bias', 'unet.downs.0.0.block2.proj.weight', 'unet.downs.0.0.block2.proj.bias', 'unet.downs.0.0.block2.norm.weight', 'unet.downs.0.0.block2.norm.bias', 'unet.downs.0.1.mlp.1.weight', 'unet.downs.0.1.mlp.1.bias', 'unet.downs.0.1.block1.proj.weight', 'unet.downs.0.1.block1.proj.bias', 'unet.downs.0.1.block1.norm.weight', 'unet.downs.0.1.block1.norm.bias', 'unet.downs.0.1.block2.proj.weight', 'unet.downs.0.1.block2.proj.bias', 'unet.downs.0.1.block2.norm.weight', 'unet.downs.0.1.block2.norm.bias', 'unet.downs.0.2.fn.fn.to_qkv.weight', 'unet.downs.0.2.fn.fn.to_out.0.weight', 'unet.downs.0.2.fn.fn.to_out.0.bias', 'unet.downs.0.2.fn.fn.to_out.1.g', 'unet.downs.0.2.fn.norm.g', 'unet.downs.0.3.weight', 'unet.downs.0.3.bias', 'unet.downs.1.0.mlp.1.weight', 'unet.downs.1.0.mlp.1.bias', 'unet.downs.1.0.block1.proj.weight', 'unet.downs.1.0.block1.proj.bias', 'unet.downs.1.0.block1.norm.weight', 'unet.downs.1.0.block1.norm.bias', 'unet.downs.1.0.block2.proj.weight', 'unet.downs.1.0.block2.proj.bias', 'unet.downs.1.0.block2.norm.weight', 'unet.downs.1.0.block2.norm.bias', 'unet.downs.1.1.mlp.1.weight', 'unet.downs.1.1.mlp.1.bias', 'unet.downs.1.1.block1.proj.weight', 'unet.downs.1.1.block1.proj.bias', 'unet.downs.1.1.block1.norm.weight', 'unet.downs.1.1.block1.norm.bias', 'unet.downs.1.1.block2.proj.weight', 'unet.downs.1.1.block2.proj.bias', 'unet.downs.1.1.block2.norm.weight', 'unet.downs.1.1.block2.norm.bias', 'unet.downs.1.2.fn.fn.to_qkv.weight', 'unet.downs.1.2.fn.fn.to_out.0.weight', 'unet.downs.1.2.fn.fn.to_out.0.bias', 'unet.downs.1.2.fn.fn.to_out.1.g', 'unet.downs.1.2.fn.norm.g', 'unet.downs.1.3.weight', 'unet.downs.1.3.bias', 'unet.downs.2.0.mlp.1.weight', 'unet.downs.2.0.mlp.1.bias', 'unet.downs.2.0.block1.proj.weight', 'unet.downs.2.0.block1.proj.bias', 'unet.downs.2.0.block1.norm.weight', 'unet.downs.2.0.block1.norm.bias', 'unet.downs.2.0.block2.proj.weight', 'unet.downs.2.0.block2.proj.bias', 'unet.downs.2.0.block2.norm.weight', 'unet.downs.2.0.block2.norm.bias', 'unet.downs.2.1.mlp.1.weight', 'unet.downs.2.1.mlp.1.bias', 'unet.downs.2.1.block1.proj.weight', 'unet.downs.2.1.block1.proj.bias', 'unet.downs.2.1.block1.norm.weight', 'unet.downs.2.1.block1.norm.bias', 'unet.downs.2.1.block2.proj.weight', 'unet.downs.2.1.block2.proj.bias', 'unet.downs.2.1.block2.norm.weight', 'unet.downs.2.1.block2.norm.bias', 'unet.downs.2.2.fn.fn.to_qkv.weight', 'unet.downs.2.2.fn.fn.to_out.0.weight', 'unet.downs.2.2.fn.fn.to_out.0.bias', 'unet.downs.2.2.fn.fn.to_out.1.g', 'unet.downs.2.2.fn.norm.g', 'unet.downs.2.3.weight', 'unet.downs.2.3.bias', 'unet.ups.0.0.mlp.1.weight', 'unet.ups.0.0.mlp.1.bias', 'unet.ups.0.0.block1.proj.weight', 'unet.ups.0.0.block1.proj.bias', 'unet.ups.0.0.block1.norm.weight', 'unet.ups.0.0.block1.norm.bias', 'unet.ups.0.0.block2.proj.weight', 'unet.ups.0.0.block2.proj.bias', 'unet.ups.0.0.block2.norm.weight', 'unet.ups.0.0.block2.norm.bias', 'unet.ups.0.0.res_conv.weight', 'unet.ups.0.0.res_conv.bias', 'unet.ups.0.1.mlp.1.weight', 'unet.ups.0.1.mlp.1.bias', 'unet.ups.0.1.block1.proj.weight', 'unet.ups.0.1.block1.proj.bias', 'unet.ups.0.1.block1.norm.weight', 'unet.ups.0.1.block1.norm.bias', 'unet.ups.0.1.block2.proj.weight', 'unet.ups.0.1.block2.proj.bias', 'unet.ups.0.1.block2.norm.weight', 'unet.ups.0.1.block2.norm.bias', 'unet.ups.0.1.res_conv.weight', 'unet.ups.0.1.res_conv.bias', 'unet.ups.0.2.fn.fn.to_qkv.weight', 'unet.ups.0.2.fn.fn.to_out.0.weight', 'unet.ups.0.2.fn.fn.to_out.0.bias', 'unet.ups.0.2.fn.fn.to_out.1.g', 'unet.ups.0.2.fn.norm.g', 'unet.ups.0.3.1.weight', 'unet.ups.0.3.1.bias', 'unet.ups.1.0.mlp.1.weight', 'unet.ups.1.0.mlp.1.bias', 'unet.ups.1.0.block1.proj.weight', 'unet.ups.1.0.block1.proj.bias', 'unet.ups.1.0.block1.norm.weight', 'unet.ups.1.0.block1.norm.bias', 'unet.ups.1.0.block2.proj.weight', 'unet.ups.1.0.block2.proj.bias', 'unet.ups.1.0.block2.norm.weight', 'unet.ups.1.0.block2.norm.bias', 'unet.ups.1.0.res_conv.weight', 'unet.ups.1.0.res_conv.bias', 'unet.ups.1.1.mlp.1.weight', 'unet.ups.1.1.mlp.1.bias', 'unet.ups.1.1.block1.proj.weight', 'unet.ups.1.1.block1.proj.bias', 'unet.ups.1.1.block1.norm.weight', 'unet.ups.1.1.block1.norm.bias', 'unet.ups.1.1.block2.proj.weight', 'unet.ups.1.1.block2.proj.bias', 'unet.ups.1.1.block2.norm.weight', 'unet.ups.1.1.block2.norm.bias', 'unet.ups.1.1.res_conv.weight', 'unet.ups.1.1.res_conv.bias', 'unet.ups.1.2.fn.fn.to_qkv.weight', 'unet.ups.1.2.fn.fn.to_out.0.weight', 'unet.ups.1.2.fn.fn.to_out.0.bias', 'unet.ups.1.2.fn.fn.to_out.1.g', 'unet.ups.1.2.fn.norm.g', 'unet.ups.1.3.1.weight', 'unet.ups.1.3.1.bias', 'unet.ups.2.0.mlp.1.weight', 'unet.ups.2.0.mlp.1.bias', 'unet.ups.2.0.block1.proj.weight', 'unet.ups.2.0.block1.proj.bias', 'unet.ups.2.0.block1.norm.weight', 'unet.ups.2.0.block1.norm.bias', 'unet.ups.2.0.block2.proj.weight', 'unet.ups.2.0.block2.proj.bias', 'unet.ups.2.0.block2.norm.weight', 'unet.ups.2.0.block2.norm.bias', 'unet.ups.2.0.res_conv.weight', 'unet.ups.2.0.res_conv.bias', 'unet.ups.2.1.mlp.1.weight', 'unet.ups.2.1.mlp.1.bias', 'unet.ups.2.1.block1.proj.weight', 'unet.ups.2.1.block1.proj.bias', 'unet.ups.2.1.block1.norm.weight', 'unet.ups.2.1.block1.norm.bias', 'unet.ups.2.1.block2.proj.weight', 'unet.ups.2.1.block2.proj.bias', 'unet.ups.2.1.block2.norm.weight', 'unet.ups.2.1.block2.norm.bias', 'unet.ups.2.1.res_conv.weight', 'unet.ups.2.1.res_conv.bias', 'unet.ups.2.2.fn.fn.to_qkv.weight', 'unet.ups.2.2.fn.fn.to_out.0.weight', 'unet.ups.2.2.fn.fn.to_out.0.bias', 'unet.ups.2.2.fn.fn.to_out.1.g', 'unet.ups.2.2.fn.norm.g', 'unet.ups.2.3.weight', 'unet.ups.2.3.bias', 'unet.mid_block1.mlp.1.weight', 'unet.mid_block1.mlp.1.bias', 'unet.mid_block1.block1.proj.weight', 'unet.mid_block1.block1.proj.bias', 'unet.mid_block1.block1.norm.weight', 'unet.mid_block1.block1.norm.bias', 'unet.mid_block1.block2.proj.weight', 'unet.mid_block1.block2.proj.bias', 'unet.mid_block1.block2.norm.weight', 'unet.mid_block1.block2.norm.bias', 'unet.mid_attn.fn.fn.to_qkv.weight', 'unet.mid_attn.fn.fn.to_out.weight', 'unet.mid_attn.fn.fn.to_out.bias', 'unet.mid_attn.fn.norm.g', 'unet.mid_block2.mlp.1.weight', 'unet.mid_block2.mlp.1.bias', 'unet.mid_block2.block1.proj.weight', 'unet.mid_block2.block1.proj.bias', 'unet.mid_block2.block1.norm.weight', 'unet.mid_block2.block1.norm.bias', 'unet.mid_block2.block2.proj.weight', 'unet.mid_block2.block2.proj.bias', 'unet.mid_block2.block2.norm.weight', 'unet.mid_block2.block2.norm.bias', 'unet.final_res_block.mlp.1.weight', 'unet.final_res_block.mlp.1.bias', 'unet.final_res_block.block1.proj.weight', 'unet.final_res_block.block1.proj.bias', 'unet.final_res_block.block1.norm.weight', 'unet.final_res_block.block1.norm.bias', 'unet.final_res_block.block2.proj.weight', 'unet.final_res_block.block2.proj.bias', 'unet.final_res_block.block2.norm.weight', 'unet.final_res_block.block2.norm.bias', 'unet.final_res_block.res_conv.weight', 'unet.final_res_block.res_conv.bias', 'unet.final_conv.weight', 'unet.final_conv.bias', 'conv_seg_new.weight', 'conv_seg_new.bias']
+2023-03-05 23:10:55,465 - mmseg - INFO - Parameters in decode_head freezed!
+2023-03-05 23:10:55,486 - mmseg - INFO - load checkpoint from local path: work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth
+2023-03-05 23:10:56,307 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: decode_head.convs.0.conv.weight, decode_head.convs.0.bn.weight, decode_head.convs.0.bn.bias, decode_head.convs.0.bn.running_mean, decode_head.convs.0.bn.running_var, decode_head.convs.0.bn.num_batches_tracked, decode_head.convs.1.conv.weight, decode_head.convs.1.bn.weight, decode_head.convs.1.bn.bias, decode_head.convs.1.bn.running_mean, decode_head.convs.1.bn.running_var, decode_head.convs.1.bn.num_batches_tracked, decode_head.convs.2.conv.weight, decode_head.convs.2.bn.weight, decode_head.convs.2.bn.bias, decode_head.convs.2.bn.running_mean, decode_head.convs.2.bn.running_var, decode_head.convs.2.bn.num_batches_tracked, decode_head.convs.3.conv.weight, decode_head.convs.3.bn.weight, decode_head.convs.3.bn.bias, decode_head.convs.3.bn.running_mean, decode_head.convs.3.bn.running_var, decode_head.convs.3.bn.num_batches_tracked, decode_head.fusion_conv.conv.weight, decode_head.fusion_conv.bn.weight, decode_head.fusion_conv.bn.bias, decode_head.fusion_conv.bn.running_mean, decode_head.fusion_conv.bn.running_var, decode_head.fusion_conv.bn.num_batches_tracked, decode_head.unet.init_conv.weight, decode_head.unet.init_conv.bias, decode_head.unet.time_mlp.1.weight, decode_head.unet.time_mlp.1.bias, decode_head.unet.time_mlp.3.weight, decode_head.unet.time_mlp.3.bias, decode_head.unet.downs.0.0.mlp.1.weight, decode_head.unet.downs.0.0.mlp.1.bias, decode_head.unet.downs.0.0.block1.proj.weight, decode_head.unet.downs.0.0.block1.proj.bias, decode_head.unet.downs.0.0.block1.norm.weight, decode_head.unet.downs.0.0.block1.norm.bias, decode_head.unet.downs.0.0.block2.proj.weight, decode_head.unet.downs.0.0.block2.proj.bias, decode_head.unet.downs.0.0.block2.norm.weight, decode_head.unet.downs.0.0.block2.norm.bias, decode_head.unet.downs.0.1.mlp.1.weight, decode_head.unet.downs.0.1.mlp.1.bias, decode_head.unet.downs.0.1.block1.proj.weight, decode_head.unet.downs.0.1.block1.proj.bias, decode_head.unet.downs.0.1.block1.norm.weight, decode_head.unet.downs.0.1.block1.norm.bias, decode_head.unet.downs.0.1.block2.proj.weight, decode_head.unet.downs.0.1.block2.proj.bias, decode_head.unet.downs.0.1.block2.norm.weight, decode_head.unet.downs.0.1.block2.norm.bias, decode_head.unet.downs.0.2.fn.fn.to_qkv.weight, decode_head.unet.downs.0.2.fn.fn.to_out.0.weight, decode_head.unet.downs.0.2.fn.fn.to_out.0.bias, decode_head.unet.downs.0.2.fn.fn.to_out.1.g, decode_head.unet.downs.0.2.fn.norm.g, decode_head.unet.downs.0.3.weight, decode_head.unet.downs.0.3.bias, decode_head.unet.downs.1.0.mlp.1.weight, decode_head.unet.downs.1.0.mlp.1.bias, decode_head.unet.downs.1.0.block1.proj.weight, decode_head.unet.downs.1.0.block1.proj.bias, decode_head.unet.downs.1.0.block1.norm.weight, decode_head.unet.downs.1.0.block1.norm.bias, decode_head.unet.downs.1.0.block2.proj.weight, decode_head.unet.downs.1.0.block2.proj.bias, decode_head.unet.downs.1.0.block2.norm.weight, decode_head.unet.downs.1.0.block2.norm.bias, decode_head.unet.downs.1.1.mlp.1.weight, decode_head.unet.downs.1.1.mlp.1.bias, decode_head.unet.downs.1.1.block1.proj.weight, decode_head.unet.downs.1.1.block1.proj.bias, decode_head.unet.downs.1.1.block1.norm.weight, decode_head.unet.downs.1.1.block1.norm.bias, decode_head.unet.downs.1.1.block2.proj.weight, decode_head.unet.downs.1.1.block2.proj.bias, decode_head.unet.downs.1.1.block2.norm.weight, decode_head.unet.downs.1.1.block2.norm.bias, decode_head.unet.downs.1.2.fn.fn.to_qkv.weight, decode_head.unet.downs.1.2.fn.fn.to_out.0.weight, decode_head.unet.downs.1.2.fn.fn.to_out.0.bias, decode_head.unet.downs.1.2.fn.fn.to_out.1.g, decode_head.unet.downs.1.2.fn.norm.g, decode_head.unet.downs.1.3.weight, decode_head.unet.downs.1.3.bias, decode_head.unet.downs.2.0.mlp.1.weight, decode_head.unet.downs.2.0.mlp.1.bias, decode_head.unet.downs.2.0.block1.proj.weight, decode_head.unet.downs.2.0.block1.proj.bias, decode_head.unet.downs.2.0.block1.norm.weight, decode_head.unet.downs.2.0.block1.norm.bias, decode_head.unet.downs.2.0.block2.proj.weight, decode_head.unet.downs.2.0.block2.proj.bias, decode_head.unet.downs.2.0.block2.norm.weight, decode_head.unet.downs.2.0.block2.norm.bias, decode_head.unet.downs.2.1.mlp.1.weight, decode_head.unet.downs.2.1.mlp.1.bias, decode_head.unet.downs.2.1.block1.proj.weight, decode_head.unet.downs.2.1.block1.proj.bias, decode_head.unet.downs.2.1.block1.norm.weight, decode_head.unet.downs.2.1.block1.norm.bias, decode_head.unet.downs.2.1.block2.proj.weight, decode_head.unet.downs.2.1.block2.proj.bias, decode_head.unet.downs.2.1.block2.norm.weight, decode_head.unet.downs.2.1.block2.norm.bias, decode_head.unet.downs.2.2.fn.fn.to_qkv.weight, decode_head.unet.downs.2.2.fn.fn.to_out.0.weight, decode_head.unet.downs.2.2.fn.fn.to_out.0.bias, decode_head.unet.downs.2.2.fn.fn.to_out.1.g, decode_head.unet.downs.2.2.fn.norm.g, decode_head.unet.downs.2.3.weight, decode_head.unet.downs.2.3.bias, decode_head.unet.ups.0.0.mlp.1.weight, decode_head.unet.ups.0.0.mlp.1.bias, decode_head.unet.ups.0.0.block1.proj.weight, decode_head.unet.ups.0.0.block1.proj.bias, decode_head.unet.ups.0.0.block1.norm.weight, decode_head.unet.ups.0.0.block1.norm.bias, decode_head.unet.ups.0.0.block2.proj.weight, decode_head.unet.ups.0.0.block2.proj.bias, decode_head.unet.ups.0.0.block2.norm.weight, decode_head.unet.ups.0.0.block2.norm.bias, decode_head.unet.ups.0.0.res_conv.weight, decode_head.unet.ups.0.0.res_conv.bias, decode_head.unet.ups.0.1.mlp.1.weight, decode_head.unet.ups.0.1.mlp.1.bias, decode_head.unet.ups.0.1.block1.proj.weight, decode_head.unet.ups.0.1.block1.proj.bias, decode_head.unet.ups.0.1.block1.norm.weight, decode_head.unet.ups.0.1.block1.norm.bias, decode_head.unet.ups.0.1.block2.proj.weight, decode_head.unet.ups.0.1.block2.proj.bias, decode_head.unet.ups.0.1.block2.norm.weight, decode_head.unet.ups.0.1.block2.norm.bias, decode_head.unet.ups.0.1.res_conv.weight, decode_head.unet.ups.0.1.res_conv.bias, decode_head.unet.ups.0.2.fn.fn.to_qkv.weight, decode_head.unet.ups.0.2.fn.fn.to_out.0.weight, decode_head.unet.ups.0.2.fn.fn.to_out.0.bias, decode_head.unet.ups.0.2.fn.fn.to_out.1.g, decode_head.unet.ups.0.2.fn.norm.g, decode_head.unet.ups.0.3.1.weight, decode_head.unet.ups.0.3.1.bias, decode_head.unet.ups.1.0.mlp.1.weight, decode_head.unet.ups.1.0.mlp.1.bias, decode_head.unet.ups.1.0.block1.proj.weight, decode_head.unet.ups.1.0.block1.proj.bias, decode_head.unet.ups.1.0.block1.norm.weight, decode_head.unet.ups.1.0.block1.norm.bias, decode_head.unet.ups.1.0.block2.proj.weight, decode_head.unet.ups.1.0.block2.proj.bias, decode_head.unet.ups.1.0.block2.norm.weight, decode_head.unet.ups.1.0.block2.norm.bias, decode_head.unet.ups.1.0.res_conv.weight, decode_head.unet.ups.1.0.res_conv.bias, decode_head.unet.ups.1.1.mlp.1.weight, decode_head.unet.ups.1.1.mlp.1.bias, decode_head.unet.ups.1.1.block1.proj.weight, decode_head.unet.ups.1.1.block1.proj.bias, decode_head.unet.ups.1.1.block1.norm.weight, decode_head.unet.ups.1.1.block1.norm.bias, decode_head.unet.ups.1.1.block2.proj.weight, decode_head.unet.ups.1.1.block2.proj.bias, decode_head.unet.ups.1.1.block2.norm.weight, decode_head.unet.ups.1.1.block2.norm.bias, decode_head.unet.ups.1.1.res_conv.weight, decode_head.unet.ups.1.1.res_conv.bias, decode_head.unet.ups.1.2.fn.fn.to_qkv.weight, decode_head.unet.ups.1.2.fn.fn.to_out.0.weight, decode_head.unet.ups.1.2.fn.fn.to_out.0.bias, decode_head.unet.ups.1.2.fn.fn.to_out.1.g, decode_head.unet.ups.1.2.fn.norm.g, decode_head.unet.ups.1.3.1.weight, decode_head.unet.ups.1.3.1.bias, decode_head.unet.ups.2.0.mlp.1.weight, decode_head.unet.ups.2.0.mlp.1.bias, decode_head.unet.ups.2.0.block1.proj.weight, decode_head.unet.ups.2.0.block1.proj.bias, decode_head.unet.ups.2.0.block1.norm.weight, decode_head.unet.ups.2.0.block1.norm.bias, decode_head.unet.ups.2.0.block2.proj.weight, decode_head.unet.ups.2.0.block2.proj.bias, decode_head.unet.ups.2.0.block2.norm.weight, decode_head.unet.ups.2.0.block2.norm.bias, decode_head.unet.ups.2.0.res_conv.weight, decode_head.unet.ups.2.0.res_conv.bias, decode_head.unet.ups.2.1.mlp.1.weight, decode_head.unet.ups.2.1.mlp.1.bias, decode_head.unet.ups.2.1.block1.proj.weight, decode_head.unet.ups.2.1.block1.proj.bias, decode_head.unet.ups.2.1.block1.norm.weight, decode_head.unet.ups.2.1.block1.norm.bias, decode_head.unet.ups.2.1.block2.proj.weight, decode_head.unet.ups.2.1.block2.proj.bias, decode_head.unet.ups.2.1.block2.norm.weight, decode_head.unet.ups.2.1.block2.norm.bias, decode_head.unet.ups.2.1.res_conv.weight, decode_head.unet.ups.2.1.res_conv.bias, decode_head.unet.ups.2.2.fn.fn.to_qkv.weight, decode_head.unet.ups.2.2.fn.fn.to_out.0.weight, decode_head.unet.ups.2.2.fn.fn.to_out.0.bias, decode_head.unet.ups.2.2.fn.fn.to_out.1.g, decode_head.unet.ups.2.2.fn.norm.g, decode_head.unet.ups.2.3.weight, decode_head.unet.ups.2.3.bias, decode_head.unet.mid_block1.mlp.1.weight, decode_head.unet.mid_block1.mlp.1.bias, decode_head.unet.mid_block1.block1.proj.weight, decode_head.unet.mid_block1.block1.proj.bias, decode_head.unet.mid_block1.block1.norm.weight, decode_head.unet.mid_block1.block1.norm.bias, decode_head.unet.mid_block1.block2.proj.weight, decode_head.unet.mid_block1.block2.proj.bias, decode_head.unet.mid_block1.block2.norm.weight, decode_head.unet.mid_block1.block2.norm.bias, decode_head.unet.mid_attn.fn.fn.to_qkv.weight, decode_head.unet.mid_attn.fn.fn.to_out.weight, decode_head.unet.mid_attn.fn.fn.to_out.bias, decode_head.unet.mid_attn.fn.norm.g, decode_head.unet.mid_block2.mlp.1.weight, decode_head.unet.mid_block2.mlp.1.bias, decode_head.unet.mid_block2.block1.proj.weight, decode_head.unet.mid_block2.block1.proj.bias, decode_head.unet.mid_block2.block1.norm.weight, decode_head.unet.mid_block2.block1.norm.bias, decode_head.unet.mid_block2.block2.proj.weight, decode_head.unet.mid_block2.block2.proj.bias, decode_head.unet.mid_block2.block2.norm.weight, decode_head.unet.mid_block2.block2.norm.bias, decode_head.unet.final_res_block.mlp.1.weight, decode_head.unet.final_res_block.mlp.1.bias, decode_head.unet.final_res_block.block1.proj.weight, decode_head.unet.final_res_block.block1.proj.bias, decode_head.unet.final_res_block.block1.norm.weight, decode_head.unet.final_res_block.block1.norm.bias, decode_head.unet.final_res_block.block2.proj.weight, decode_head.unet.final_res_block.block2.proj.bias, decode_head.unet.final_res_block.block2.norm.weight, decode_head.unet.final_res_block.block2.norm.bias, decode_head.unet.final_res_block.res_conv.weight, decode_head.unet.final_res_block.res_conv.bias, decode_head.unet.final_conv.weight, decode_head.unet.final_conv.bias, decode_head.conv_seg_new.weight, decode_head.conv_seg_new.bias, decode_head.embed.weight
+
+2023-03-05 23:10:56,324 - mmseg - INFO - load checkpoint from local path: work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth
+2023-03-05 23:10:56,771 - mmseg - WARNING - The model and loaded state dict do not match exactly
+
+unexpected key in source state_dict: backbone.layers.0.0.projection.weight, backbone.layers.0.0.projection.bias, backbone.layers.0.0.norm.weight, backbone.layers.0.0.norm.bias, backbone.layers.0.1.0.norm1.weight, backbone.layers.0.1.0.norm1.bias, backbone.layers.0.1.0.attn.attn.in_proj_weight, backbone.layers.0.1.0.attn.attn.in_proj_bias, backbone.layers.0.1.0.attn.attn.out_proj.weight, backbone.layers.0.1.0.attn.attn.out_proj.bias, backbone.layers.0.1.0.attn.sr.weight, backbone.layers.0.1.0.attn.sr.bias, backbone.layers.0.1.0.attn.norm.weight, backbone.layers.0.1.0.attn.norm.bias, backbone.layers.0.1.0.norm2.weight, backbone.layers.0.1.0.norm2.bias, backbone.layers.0.1.0.ffn.layers.0.weight, backbone.layers.0.1.0.ffn.layers.0.bias, backbone.layers.0.1.0.ffn.layers.1.weight, backbone.layers.0.1.0.ffn.layers.1.bias, backbone.layers.0.1.0.ffn.layers.4.weight, backbone.layers.0.1.0.ffn.layers.4.bias, backbone.layers.0.1.1.norm1.weight, backbone.layers.0.1.1.norm1.bias, backbone.layers.0.1.1.attn.attn.in_proj_weight, backbone.layers.0.1.1.attn.attn.in_proj_bias, backbone.layers.0.1.1.attn.attn.out_proj.weight, backbone.layers.0.1.1.attn.attn.out_proj.bias, backbone.layers.0.1.1.attn.sr.weight, backbone.layers.0.1.1.attn.sr.bias, backbone.layers.0.1.1.attn.norm.weight, backbone.layers.0.1.1.attn.norm.bias, backbone.layers.0.1.1.norm2.weight, backbone.layers.0.1.1.norm2.bias, backbone.layers.0.1.1.ffn.layers.0.weight, backbone.layers.0.1.1.ffn.layers.0.bias, backbone.layers.0.1.1.ffn.layers.1.weight, backbone.layers.0.1.1.ffn.layers.1.bias, backbone.layers.0.1.1.ffn.layers.4.weight, backbone.layers.0.1.1.ffn.layers.4.bias, backbone.layers.0.1.2.norm1.weight, backbone.layers.0.1.2.norm1.bias, backbone.layers.0.1.2.attn.attn.in_proj_weight, backbone.layers.0.1.2.attn.attn.in_proj_bias, backbone.layers.0.1.2.attn.attn.out_proj.weight, backbone.layers.0.1.2.attn.attn.out_proj.bias, backbone.layers.0.1.2.attn.sr.weight, backbone.layers.0.1.2.attn.sr.bias, backbone.layers.0.1.2.attn.norm.weight, backbone.layers.0.1.2.attn.norm.bias, backbone.layers.0.1.2.norm2.weight, backbone.layers.0.1.2.norm2.bias, backbone.layers.0.1.2.ffn.layers.0.weight, backbone.layers.0.1.2.ffn.layers.0.bias, backbone.layers.0.1.2.ffn.layers.1.weight, backbone.layers.0.1.2.ffn.layers.1.bias, backbone.layers.0.1.2.ffn.layers.4.weight, backbone.layers.0.1.2.ffn.layers.4.bias, backbone.layers.0.2.weight, backbone.layers.0.2.bias, backbone.layers.1.0.projection.weight, backbone.layers.1.0.projection.bias, backbone.layers.1.0.norm.weight, backbone.layers.1.0.norm.bias, backbone.layers.1.1.0.norm1.weight, backbone.layers.1.1.0.norm1.bias, backbone.layers.1.1.0.attn.attn.in_proj_weight, backbone.layers.1.1.0.attn.attn.in_proj_bias, backbone.layers.1.1.0.attn.attn.out_proj.weight, backbone.layers.1.1.0.attn.attn.out_proj.bias, backbone.layers.1.1.0.attn.sr.weight, backbone.layers.1.1.0.attn.sr.bias, backbone.layers.1.1.0.attn.norm.weight, backbone.layers.1.1.0.attn.norm.bias, backbone.layers.1.1.0.norm2.weight, backbone.layers.1.1.0.norm2.bias, backbone.layers.1.1.0.ffn.layers.0.weight, backbone.layers.1.1.0.ffn.layers.0.bias, backbone.layers.1.1.0.ffn.layers.1.weight, backbone.layers.1.1.0.ffn.layers.1.bias, backbone.layers.1.1.0.ffn.layers.4.weight, backbone.layers.1.1.0.ffn.layers.4.bias, backbone.layers.1.1.1.norm1.weight, backbone.layers.1.1.1.norm1.bias, backbone.layers.1.1.1.attn.attn.in_proj_weight, backbone.layers.1.1.1.attn.attn.in_proj_bias, backbone.layers.1.1.1.attn.attn.out_proj.weight, backbone.layers.1.1.1.attn.attn.out_proj.bias, backbone.layers.1.1.1.attn.sr.weight, backbone.layers.1.1.1.attn.sr.bias, backbone.layers.1.1.1.attn.norm.weight, backbone.layers.1.1.1.attn.norm.bias, backbone.layers.1.1.1.norm2.weight, backbone.layers.1.1.1.norm2.bias, backbone.layers.1.1.1.ffn.layers.0.weight, backbone.layers.1.1.1.ffn.layers.0.bias, backbone.layers.1.1.1.ffn.layers.1.weight, backbone.layers.1.1.1.ffn.layers.1.bias, backbone.layers.1.1.1.ffn.layers.4.weight, backbone.layers.1.1.1.ffn.layers.4.bias, backbone.layers.1.1.2.norm1.weight, backbone.layers.1.1.2.norm1.bias, backbone.layers.1.1.2.attn.attn.in_proj_weight, backbone.layers.1.1.2.attn.attn.in_proj_bias, backbone.layers.1.1.2.attn.attn.out_proj.weight, backbone.layers.1.1.2.attn.attn.out_proj.bias, backbone.layers.1.1.2.attn.sr.weight, backbone.layers.1.1.2.attn.sr.bias, backbone.layers.1.1.2.attn.norm.weight, backbone.layers.1.1.2.attn.norm.bias, backbone.layers.1.1.2.norm2.weight, backbone.layers.1.1.2.norm2.bias, backbone.layers.1.1.2.ffn.layers.0.weight, backbone.layers.1.1.2.ffn.layers.0.bias, backbone.layers.1.1.2.ffn.layers.1.weight, backbone.layers.1.1.2.ffn.layers.1.bias, backbone.layers.1.1.2.ffn.layers.4.weight, backbone.layers.1.1.2.ffn.layers.4.bias, backbone.layers.1.1.3.norm1.weight, backbone.layers.1.1.3.norm1.bias, backbone.layers.1.1.3.attn.attn.in_proj_weight, backbone.layers.1.1.3.attn.attn.in_proj_bias, backbone.layers.1.1.3.attn.attn.out_proj.weight, backbone.layers.1.1.3.attn.attn.out_proj.bias, backbone.layers.1.1.3.attn.sr.weight, backbone.layers.1.1.3.attn.sr.bias, backbone.layers.1.1.3.attn.norm.weight, backbone.layers.1.1.3.attn.norm.bias, backbone.layers.1.1.3.norm2.weight, backbone.layers.1.1.3.norm2.bias, backbone.layers.1.1.3.ffn.layers.0.weight, backbone.layers.1.1.3.ffn.layers.0.bias, backbone.layers.1.1.3.ffn.layers.1.weight, backbone.layers.1.1.3.ffn.layers.1.bias, backbone.layers.1.1.3.ffn.layers.4.weight, backbone.layers.1.1.3.ffn.layers.4.bias, backbone.layers.1.2.weight, backbone.layers.1.2.bias, backbone.layers.2.0.projection.weight, backbone.layers.2.0.projection.bias, backbone.layers.2.0.norm.weight, backbone.layers.2.0.norm.bias, backbone.layers.2.1.0.norm1.weight, backbone.layers.2.1.0.norm1.bias, backbone.layers.2.1.0.attn.attn.in_proj_weight, backbone.layers.2.1.0.attn.attn.in_proj_bias, backbone.layers.2.1.0.attn.attn.out_proj.weight, backbone.layers.2.1.0.attn.attn.out_proj.bias, backbone.layers.2.1.0.attn.sr.weight, backbone.layers.2.1.0.attn.sr.bias, backbone.layers.2.1.0.attn.norm.weight, backbone.layers.2.1.0.attn.norm.bias, backbone.layers.2.1.0.norm2.weight, backbone.layers.2.1.0.norm2.bias, backbone.layers.2.1.0.ffn.layers.0.weight, backbone.layers.2.1.0.ffn.layers.0.bias, backbone.layers.2.1.0.ffn.layers.1.weight, backbone.layers.2.1.0.ffn.layers.1.bias, backbone.layers.2.1.0.ffn.layers.4.weight, backbone.layers.2.1.0.ffn.layers.4.bias, backbone.layers.2.1.1.norm1.weight, backbone.layers.2.1.1.norm1.bias, backbone.layers.2.1.1.attn.attn.in_proj_weight, backbone.layers.2.1.1.attn.attn.in_proj_bias, backbone.layers.2.1.1.attn.attn.out_proj.weight, backbone.layers.2.1.1.attn.attn.out_proj.bias, backbone.layers.2.1.1.attn.sr.weight, backbone.layers.2.1.1.attn.sr.bias, backbone.layers.2.1.1.attn.norm.weight, backbone.layers.2.1.1.attn.norm.bias, backbone.layers.2.1.1.norm2.weight, backbone.layers.2.1.1.norm2.bias, backbone.layers.2.1.1.ffn.layers.0.weight, backbone.layers.2.1.1.ffn.layers.0.bias, backbone.layers.2.1.1.ffn.layers.1.weight, backbone.layers.2.1.1.ffn.layers.1.bias, backbone.layers.2.1.1.ffn.layers.4.weight, backbone.layers.2.1.1.ffn.layers.4.bias, backbone.layers.2.1.2.norm1.weight, backbone.layers.2.1.2.norm1.bias, backbone.layers.2.1.2.attn.attn.in_proj_weight, backbone.layers.2.1.2.attn.attn.in_proj_bias, backbone.layers.2.1.2.attn.attn.out_proj.weight, backbone.layers.2.1.2.attn.attn.out_proj.bias, backbone.layers.2.1.2.attn.sr.weight, backbone.layers.2.1.2.attn.sr.bias, backbone.layers.2.1.2.attn.norm.weight, backbone.layers.2.1.2.attn.norm.bias, backbone.layers.2.1.2.norm2.weight, backbone.layers.2.1.2.norm2.bias, backbone.layers.2.1.2.ffn.layers.0.weight, backbone.layers.2.1.2.ffn.layers.0.bias, backbone.layers.2.1.2.ffn.layers.1.weight, backbone.layers.2.1.2.ffn.layers.1.bias, backbone.layers.2.1.2.ffn.layers.4.weight, backbone.layers.2.1.2.ffn.layers.4.bias, backbone.layers.2.1.3.norm1.weight, backbone.layers.2.1.3.norm1.bias, backbone.layers.2.1.3.attn.attn.in_proj_weight, backbone.layers.2.1.3.attn.attn.in_proj_bias, backbone.layers.2.1.3.attn.attn.out_proj.weight, backbone.layers.2.1.3.attn.attn.out_proj.bias, backbone.layers.2.1.3.attn.sr.weight, backbone.layers.2.1.3.attn.sr.bias, backbone.layers.2.1.3.attn.norm.weight, backbone.layers.2.1.3.attn.norm.bias, backbone.layers.2.1.3.norm2.weight, backbone.layers.2.1.3.norm2.bias, backbone.layers.2.1.3.ffn.layers.0.weight, backbone.layers.2.1.3.ffn.layers.0.bias, backbone.layers.2.1.3.ffn.layers.1.weight, backbone.layers.2.1.3.ffn.layers.1.bias, backbone.layers.2.1.3.ffn.layers.4.weight, backbone.layers.2.1.3.ffn.layers.4.bias, backbone.layers.2.1.4.norm1.weight, backbone.layers.2.1.4.norm1.bias, backbone.layers.2.1.4.attn.attn.in_proj_weight, backbone.layers.2.1.4.attn.attn.in_proj_bias, backbone.layers.2.1.4.attn.attn.out_proj.weight, backbone.layers.2.1.4.attn.attn.out_proj.bias, backbone.layers.2.1.4.attn.sr.weight, backbone.layers.2.1.4.attn.sr.bias, backbone.layers.2.1.4.attn.norm.weight, backbone.layers.2.1.4.attn.norm.bias, backbone.layers.2.1.4.norm2.weight, backbone.layers.2.1.4.norm2.bias, backbone.layers.2.1.4.ffn.layers.0.weight, backbone.layers.2.1.4.ffn.layers.0.bias, backbone.layers.2.1.4.ffn.layers.1.weight, backbone.layers.2.1.4.ffn.layers.1.bias, backbone.layers.2.1.4.ffn.layers.4.weight, backbone.layers.2.1.4.ffn.layers.4.bias, backbone.layers.2.1.5.norm1.weight, backbone.layers.2.1.5.norm1.bias, backbone.layers.2.1.5.attn.attn.in_proj_weight, backbone.layers.2.1.5.attn.attn.in_proj_bias, backbone.layers.2.1.5.attn.attn.out_proj.weight, backbone.layers.2.1.5.attn.attn.out_proj.bias, backbone.layers.2.1.5.attn.sr.weight, backbone.layers.2.1.5.attn.sr.bias, backbone.layers.2.1.5.attn.norm.weight, backbone.layers.2.1.5.attn.norm.bias, backbone.layers.2.1.5.norm2.weight, backbone.layers.2.1.5.norm2.bias, backbone.layers.2.1.5.ffn.layers.0.weight, backbone.layers.2.1.5.ffn.layers.0.bias, backbone.layers.2.1.5.ffn.layers.1.weight, backbone.layers.2.1.5.ffn.layers.1.bias, backbone.layers.2.1.5.ffn.layers.4.weight, backbone.layers.2.1.5.ffn.layers.4.bias, backbone.layers.2.2.weight, backbone.layers.2.2.bias, backbone.layers.3.0.projection.weight, backbone.layers.3.0.projection.bias, backbone.layers.3.0.norm.weight, backbone.layers.3.0.norm.bias, backbone.layers.3.1.0.norm1.weight, backbone.layers.3.1.0.norm1.bias, backbone.layers.3.1.0.attn.attn.in_proj_weight, backbone.layers.3.1.0.attn.attn.in_proj_bias, backbone.layers.3.1.0.attn.attn.out_proj.weight, backbone.layers.3.1.0.attn.attn.out_proj.bias, backbone.layers.3.1.0.norm2.weight, backbone.layers.3.1.0.norm2.bias, backbone.layers.3.1.0.ffn.layers.0.weight, backbone.layers.3.1.0.ffn.layers.0.bias, backbone.layers.3.1.0.ffn.layers.1.weight, backbone.layers.3.1.0.ffn.layers.1.bias, backbone.layers.3.1.0.ffn.layers.4.weight, backbone.layers.3.1.0.ffn.layers.4.bias, backbone.layers.3.1.1.norm1.weight, backbone.layers.3.1.1.norm1.bias, backbone.layers.3.1.1.attn.attn.in_proj_weight, backbone.layers.3.1.1.attn.attn.in_proj_bias, backbone.layers.3.1.1.attn.attn.out_proj.weight, backbone.layers.3.1.1.attn.attn.out_proj.bias, backbone.layers.3.1.1.norm2.weight, backbone.layers.3.1.1.norm2.bias, backbone.layers.3.1.1.ffn.layers.0.weight, backbone.layers.3.1.1.ffn.layers.0.bias, backbone.layers.3.1.1.ffn.layers.1.weight, backbone.layers.3.1.1.ffn.layers.1.bias, backbone.layers.3.1.1.ffn.layers.4.weight, backbone.layers.3.1.1.ffn.layers.4.bias, backbone.layers.3.1.2.norm1.weight, backbone.layers.3.1.2.norm1.bias, backbone.layers.3.1.2.attn.attn.in_proj_weight, backbone.layers.3.1.2.attn.attn.in_proj_bias, backbone.layers.3.1.2.attn.attn.out_proj.weight, backbone.layers.3.1.2.attn.attn.out_proj.bias, backbone.layers.3.1.2.norm2.weight, backbone.layers.3.1.2.norm2.bias, backbone.layers.3.1.2.ffn.layers.0.weight, backbone.layers.3.1.2.ffn.layers.0.bias, backbone.layers.3.1.2.ffn.layers.1.weight, backbone.layers.3.1.2.ffn.layers.1.bias, backbone.layers.3.1.2.ffn.layers.4.weight, backbone.layers.3.1.2.ffn.layers.4.bias, backbone.layers.3.2.weight, backbone.layers.3.2.bias
+
+missing keys in source state_dict: log_cumprod_at, log_cumprod_bt, log_at, log_bt
+
+2023-03-05 23:10:56,795 - mmseg - INFO - EncoderDecoderDiffusion(
+  (backbone): MixVisionTransformerCustomInitWeights(
+    (layers): ModuleList(
+      (0): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(3, 64, kernel_size=(7, 7), stride=(4, 4), padding=(3, 3))
+          (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=64, out_features=64, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(64, 64, kernel_size=(8, 8), stride=(8, 8))
+              (norm): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=256)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((64,), eps=1e-06, elementwise_affine=True)
+      )
+      (1): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=128, out_features=128, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(128, 128, kernel_size=(4, 4), stride=(4, 4))
+              (norm): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=512)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((128,), eps=1e-06, elementwise_affine=True)
+      )
+      (2): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(128, 320, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (3): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (4): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (5): TransformerEncoderLayer(
+            (norm1): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=320, out_features=320, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+              (sr): Conv2d(320, 320, kernel_size=(2, 2), stride=(2, 2))
+              (norm): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            )
+            (norm2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(1280, 1280, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1280)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((320,), eps=1e-06, elementwise_affine=True)
+      )
+      (3): ModuleList(
+        (0): PatchEmbed(
+          (projection): Conv2d(320, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
+          (norm): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+        )
+        (1): ModuleList(
+          (0): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (1): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+          (2): TransformerEncoderLayer(
+            (norm1): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (attn): EfficientMultiheadAttention(
+              (attn): MultiheadAttention(
+                (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
+              )
+              (proj_drop): Dropout(p=0.0, inplace=False)
+              (dropout_layer): DropPath()
+            )
+            (norm2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+            (ffn): MixFFN(
+              (activate): GELU(approximate='none')
+              (layers): Sequential(
+                (0): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
+                (1): Conv2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=2048)
+                (2): GELU(approximate='none')
+                (3): Dropout(p=0.0, inplace=False)
+                (4): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
+                (5): Dropout(p=0.0, inplace=False)
+              )
+              (dropout_layer): DropPath()
+            )
+          )
+        )
+        (2): LayerNorm((512,), eps=1e-06, elementwise_affine=True)
+      )
+    )
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'}
+  (decode_head): SegformerHeadUnetFCHeadMultiStepCE(
+    input_transform=multiple_select, ignore_index=0, align_corners=False
+    (loss_decode): CrossEntropyLoss(avg_non_ignore=False)
+    (conv_seg): None
+    (dropout): Dropout2d(p=0.1, inplace=False)
+    (convs): ModuleList(
+      (0): ConvModule(
+        (conv): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (1): ConvModule(
+        (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (2): ConvModule(
+        (conv): Conv2d(320, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+      (3): ConvModule(
+        (conv): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+        (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+        (activate): ReLU(inplace=True)
+      )
+    )
+    (fusion_conv): ConvModule(
+      (conv): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
+      (bn): SyncBatchNorm(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
+      (activate): ReLU(inplace=True)
+    )
+    (unet): Unet(
+      (init_conv): Conv2d(272, 128, kernel_size=(7, 7), stride=(1, 1), padding=(3, 3))
+      (time_mlp): Sequential(
+        (0): SinusoidalPosEmb()
+        (1): Linear(in_features=128, out_features=512, bias=True)
+        (2): GELU(approximate='none')
+        (3): Linear(in_features=512, out_features=512, bias=True)
+      )
+      (downs): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Identity()
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (ups): ModuleList(
+        (0): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (1): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Sequential(
+            (0): Upsample(scale_factor=2.0, mode=nearest)
+            (1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          )
+        )
+        (2): ModuleList(
+          (0): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (1): ResnetBlock(
+            (mlp): Sequential(
+              (0): SiLU()
+              (1): Linear(in_features=512, out_features=256, bias=True)
+            )
+            (block1): Block(
+              (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (block2): Block(
+              (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+              (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+              (act): SiLU()
+            )
+            (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (2): Residual(
+            (fn): PreNorm(
+              (fn): LinearAttention(
+                (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+                (to_out): Sequential(
+                  (0): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+                  (1): LayerNorm()
+                )
+              )
+              (norm): LayerNorm()
+            )
+          )
+          (3): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+        )
+      )
+      (mid_block1): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (mid_attn): Residual(
+        (fn): PreNorm(
+          (fn): Attention(
+            (to_qkv): Conv2d(128, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
+            (to_out): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
+          )
+          (norm): LayerNorm()
+        )
+      )
+      (mid_block2): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Identity()
+      )
+      (final_res_block): ResnetBlock(
+        (mlp): Sequential(
+          (0): SiLU()
+          (1): Linear(in_features=512, out_features=256, bias=True)
+        )
+        (block1): Block(
+          (proj): WeightStandardizedConv2d(256, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (block2): Block(
+          (proj): WeightStandardizedConv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
+          (norm): GroupNorm(8, 128, eps=1e-05, affine=True)
+          (act): SiLU()
+        )
+        (res_conv): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
+      )
+      (final_conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1))
+    )
+    (conv_seg_new): Conv2d(256, 151, kernel_size=(1, 1), stride=(1, 1))
+    (embed): Embedding(151, 16)
+  )
+  init_cfg={'type': 'Pretrained', 'checkpoint': 'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'}
+)
+2023-03-05 23:10:57,286 - mmseg - INFO - Loaded 20210 images
+2023-03-05 23:11:00,862 - mmseg - INFO - Loaded 2000 images
+2023-03-05 23:11:00,864 - mmseg - INFO - Start running, host: laizeqiang@SH-IDC1-10-140-37-110, work_dir: /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce
+2023-03-05 23:11:00,865 - mmseg - INFO - Hooks will be executed in the following order:
+before_run:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(49          ) ConstantMomentumEMAHook            
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_epoch:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_train_iter:
+(VERY_HIGH   ) StepLrUpdaterHook                  
+(49          ) ConstantMomentumEMAHook            
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+ -------------------- 
+after_train_iter:
+(ABOVE_NORMAL) OptimizerHook                      
+(49          ) ConstantMomentumEMAHook            
+(NORMAL      ) CheckpointHook                     
+(LOW         ) IterTimerHook                      
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_train_epoch:
+(NORMAL      ) CheckpointHook                     
+(LOW         ) DistEvalHookMultiSteps             
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_epoch:
+(LOW         ) IterTimerHook                      
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+before_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_iter:
+(LOW         ) IterTimerHook                      
+ -------------------- 
+after_val_epoch:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+after_run:
+(VERY_LOW    ) TextLoggerHook                     
+ -------------------- 
+2023-03-05 23:11:00,865 - mmseg - INFO - workflow: [('train', 1)], max: 160000 iters
+2023-03-05 23:11:00,901 - mmseg - INFO - Checkpoints will be saved to /mnt/petrelfs/laizeqiang/mmseg-baseline/work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce by HardDiskBackend.
+2023-03-05 23:11:25,138 - mmseg - INFO - Swap parameters (before train) before iter [1]

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/20230305_231050.log.json

@@ -0,0 +1 @@
+{"env_info": "sys.platform: linux\nPython: 3.7.16 (default, Jan 17 2023, 22:20:44) [GCC 11.2.0]\nCUDA available: True\nGPU 0,1,2,3,4,5,6,7: NVIDIA A100-SXM4-80GB\nCUDA_HOME: /mnt/petrelfs/laizeqiang/miniconda3/envs/torch\nNVCC: Cuda compilation tools, release 11.6, V11.6.124\nGCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-44)\nPyTorch: 1.13.1\nPyTorch compiling details: PyTorch built with:\n  - GCC 9.3\n  - C++ Version: 201402\n  - Intel(R) oneAPI Math Kernel Library Version 2021.4-Product Build 20210904 for Intel(R) 64 architecture applications\n  - Intel(R) MKL-DNN v2.6.0 (Git Hash 52b5f107dd9cf10910aaa19cb47f3abf9b349815)\n  - OpenMP 201511 (a.k.a. OpenMP 4.5)\n  - LAPACK is enabled (usually provided by MKL)\n  - NNPACK is enabled\n  - CPU capability usage: AVX2\n  - CUDA Runtime 11.6\n  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_37,code=compute_37\n  - CuDNN 8.3.2  (built against CUDA 11.5)\n  - Magma 2.6.1\n  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.6, CUDNN_VERSION=8.3.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.13.1, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, \n\nTorchVision: 0.14.1\nOpenCV: 4.7.0\nMMCV: 1.7.1\nMMCV Compiler: GCC 9.3\nMMCV CUDA Compiler: 11.6\nMMSegmentation: 0.30.0+6db5ece", "seed": 1580901347, "exp_name": "ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce.py", "mmseg_version": "0.30.0+6db5ece", "config": "norm_cfg = dict(type='SyncBN', requires_grad=True)\ncheckpoint = 'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'\nmodel = dict(\n    type='EncoderDecoderDiffusion',\n    freeze_parameters=['backbone', 'decode_head'],\n    pretrained=\n    'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',\n    backbone=dict(\n        type='MixVisionTransformerCustomInitWeights',\n        in_channels=3,\n        embed_dims=64,\n        num_stages=4,\n        num_layers=[3, 4, 6, 3],\n        num_heads=[1, 2, 5, 8],\n        patch_sizes=[7, 3, 3, 3],\n        sr_ratios=[8, 4, 2, 1],\n        out_indices=(0, 1, 2, 3),\n        mlp_ratio=4,\n        qkv_bias=True,\n        drop_rate=0.0,\n        attn_drop_rate=0.0,\n        drop_path_rate=0.1,\n        pretrained=\n        'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'\n    ),\n    decode_head=dict(\n        type='SegformerHeadUnetFCHeadMultiStepCE',\n        pretrained=\n        'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',\n        dim=128,\n        out_dim=256,\n        unet_channels=272,\n        dim_mults=[1, 1, 1],\n        cat_embedding_dim=16,\n        diffusion_timesteps=100,\n        collect_timesteps=[0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 99],\n        in_channels=[64, 128, 320, 512],\n        in_index=[0, 1, 2, 3],\n        channels=256,\n        dropout_ratio=0.1,\n        num_classes=151,\n        norm_cfg=dict(type='SyncBN', requires_grad=True),\n        align_corners=False,\n        ignore_index=0,\n        loss_decode=dict(\n            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.1)),\n    train_cfg=dict(),\n    test_cfg=dict(mode='whole'))\ndataset_type = 'ADE20K151Dataset'\ndata_root = 'data/ade/ADEChallengeData2016'\nimg_norm_cfg = dict(\n    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)\ncrop_size = (512, 512)\ntrain_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(type='LoadAnnotations', reduce_zero_label=False),\n    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n    dict(type='RandomFlip', prob=0.5),\n    dict(type='PhotoMetricDistortion'),\n    dict(\n        type='Normalize',\n        mean=[123.675, 116.28, 103.53],\n        std=[58.395, 57.12, 57.375],\n        to_rgb=True),\n    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n    dict(type='DefaultFormatBundle'),\n    dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n]\ntest_pipeline = [\n    dict(type='LoadImageFromFile'),\n    dict(\n        type='MultiScaleFlipAug',\n        img_scale=(2048, 512),\n        flip=False,\n        transforms=[\n            dict(type='Resize', keep_ratio=True),\n            dict(type='RandomFlip'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n            dict(type='ImageToTensor', keys=['img']),\n            dict(type='Collect', keys=['img'])\n        ])\n]\ndata = dict(\n    samples_per_gpu=4,\n    workers_per_gpu=4,\n    train=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/training',\n        ann_dir='annotations/training',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(type='LoadAnnotations', reduce_zero_label=False),\n            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),\n            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),\n            dict(type='RandomFlip', prob=0.5),\n            dict(type='PhotoMetricDistortion'),\n            dict(\n                type='Normalize',\n                mean=[123.675, 116.28, 103.53],\n                std=[58.395, 57.12, 57.375],\n                to_rgb=True),\n            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),\n            dict(type='DefaultFormatBundle'),\n            dict(type='Collect', keys=['img', 'gt_semantic_seg'])\n        ]),\n    val=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]),\n    test=dict(\n        type='ADE20K151Dataset',\n        data_root='data/ade/ADEChallengeData2016',\n        img_dir='images/validation',\n        ann_dir='annotations/validation',\n        pipeline=[\n            dict(type='LoadImageFromFile'),\n            dict(\n                type='MultiScaleFlipAug',\n                img_scale=(2048, 512),\n                flip=False,\n                transforms=[\n                    dict(type='Resize', keep_ratio=True),\n                    dict(type='RandomFlip'),\n                    dict(\n                        type='Normalize',\n                        mean=[123.675, 116.28, 103.53],\n                        std=[58.395, 57.12, 57.375],\n                        to_rgb=True),\n                    dict(\n                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),\n                    dict(type='ImageToTensor', keys=['img']),\n                    dict(type='Collect', keys=['img'])\n                ])\n        ]))\nlog_config = dict(\n    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])\ndist_params = dict(backend='nccl')\nlog_level = 'INFO'\nload_from = None\nresume_from = None\nworkflow = [('train', 1)]\ncudnn_benchmark = True\noptimizer = dict(\n    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)\noptimizer_config = dict()\nlr_config = dict(\n    policy='step',\n    warmup='linear',\n    warmup_iters=1000,\n    warmup_ratio=1e-06,\n    step=20000,\n    gamma=0.5,\n    min_lr=1e-06,\n    by_epoch=False)\nrunner = dict(type='IterBasedRunner', max_iters=160000)\ncheckpoint_config = dict(by_epoch=False, interval=16000, max_keep_ckpts=1)\nevaluation = dict(\n    interval=16000, metric='mIoU', pre_eval=True, save_best='mIoU')\ncustom_hooks = [\n    dict(\n        type='ConstantMomentumEMAHook',\n        momentum=0.01,\n        interval=25,\n        eval_interval=16000,\n        auto_resume=True,\n        priority=49)\n]\nwork_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce'\ngpu_ids = range(0, 8)\nauto_resume = True\ndevice = 'cuda'\nseed = 1580901347\n", "CLASSES": ["background", "wall", "building", "sky", "floor", "tree", "ceiling", "road", "bed ", "windowpane", "grass", "cabinet", "sidewalk", "person", "earth", "door", "table", "mountain", "plant", "curtain", "chair", "car", "water", "painting", "sofa", "shelf", "house", "sea", "mirror", "rug", "field", "armchair", "seat", "fence", "desk", "rock", "wardrobe", "lamp", "bathtub", "railing", "cushion", "base", "box", "column", "signboard", "chest of drawers", "counter", "sand", "sink", "skyscraper", "fireplace", "refrigerator", "grandstand", "path", "stairs", "runway", "case", "pool table", "pillow", "screen door", "stairway", "river", "bridge", "bookcase", "blind", "coffee table", "toilet", "flower", "book", "hill", "bench", "countertop", "stove", "palm", "kitchen island", "computer", "swivel chair", "boat", "bar", "arcade machine", "hovel", "bus", "towel", "light", "truck", "tower", "chandelier", "awning", "streetlight", "booth", "television receiver", "airplane", "dirt track", "apparel", "pole", "land", "bannister", "escalator", "ottoman", "bottle", "buffet", "poster", "stage", "van", "ship", "fountain", "conveyer belt", "canopy", "washer", "plaything", "swimming pool", "stool", "barrel", "basket", "waterfall", "tent", "bag", "minibike", "cradle", "oven", "ball", "food", "step", "tank", "trade name", "microwave", "pot", "animal", "bicycle", "lake", "dishwasher", "screen", "blanket", "sculpture", "hood", "sconce", "vase", "traffic light", "tray", "ashcan", "fan", "pier", "crt screen", "plate", "monitor", "bulletin board", "shower", "radiator", "glass", "clock", "flag"], "PALETTE": [[0, 0, 0], [120, 120, 120], [180, 120, 120], [6, 230, 230], [80, 50, 50], [4, 200, 3], [120, 120, 80], [140, 140, 140], [204, 5, 255], [230, 230, 230], [4, 250, 7], [224, 5, 255], [235, 255, 7], [150, 5, 61], [120, 120, 70], [8, 255, 51], [255, 6, 82], [143, 255, 140], [204, 255, 4], [255, 51, 7], [204, 70, 3], [0, 102, 200], [61, 230, 250], [255, 6, 51], [11, 102, 255], [255, 7, 71], [255, 9, 224], [9, 7, 230], [220, 220, 220], [255, 9, 92], [112, 9, 255], [8, 255, 214], [7, 255, 224], [255, 184, 6], [10, 255, 71], [255, 41, 10], [7, 255, 255], [224, 255, 8], [102, 8, 255], [255, 61, 6], [255, 194, 7], [255, 122, 8], [0, 255, 20], [255, 8, 41], [255, 5, 153], [6, 51, 255], [235, 12, 255], [160, 150, 20], [0, 163, 255], [140, 140, 140], [250, 10, 15], [20, 255, 0], [31, 255, 0], [255, 31, 0], [255, 224, 0], [153, 255, 0], [0, 0, 255], [255, 71, 0], [0, 235, 255], [0, 173, 255], [31, 0, 255], [11, 200, 200], [255, 82, 0], [0, 255, 245], [0, 61, 255], [0, 255, 112], [0, 255, 133], [255, 0, 0], [255, 163, 0], [255, 102, 0], [194, 255, 0], [0, 143, 255], [51, 255, 0], [0, 82, 255], [0, 255, 41], [0, 255, 173], [10, 0, 255], [173, 255, 0], [0, 255, 153], [255, 92, 0], [255, 0, 255], [255, 0, 245], [255, 0, 102], [255, 173, 0], [255, 0, 20], [255, 184, 184], [0, 31, 255], [0, 255, 61], [0, 71, 255], [255, 0, 204], [0, 255, 194], [0, 255, 82], [0, 10, 255], [0, 112, 255], [51, 0, 255], [0, 194, 255], [0, 122, 255], [0, 255, 163], [255, 153, 0], [0, 255, 10], [255, 112, 0], [143, 255, 0], [82, 0, 255], [163, 255, 0], [255, 235, 0], [8, 184, 170], [133, 0, 255], [0, 255, 92], [184, 0, 255], [255, 0, 31], [0, 184, 255], [0, 214, 255], [255, 0, 112], [92, 255, 0], [0, 224, 255], [112, 224, 255], [70, 184, 160], [163, 0, 255], [153, 0, 255], [71, 255, 0], [255, 0, 163], [255, 204, 0], [255, 0, 143], [0, 255, 235], [133, 255, 0], [255, 0, 235], [245, 0, 255], [255, 0, 122], [255, 245, 0], [10, 190, 212], [214, 255, 0], [0, 204, 255], [20, 0, 255], [255, 255, 0], [0, 153, 255], [0, 41, 255], [0, 255, 204], [41, 0, 255], [41, 255, 0], [173, 0, 255], [0, 245, 255], [71, 0, 255], [122, 0, 255], [0, 255, 184], [0, 92, 255], [184, 255, 0], [0, 133, 255], [255, 214, 0], [25, 194, 194], [102, 255, 0], [92, 0, 255]], "hook_msgs": {}}

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce.py

@@ -0,0 +1,195 @@
+norm_cfg = dict(type='SyncBN', requires_grad=True)
+checkpoint = 'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth'
+model = dict(
+    type='EncoderDecoderDiffusion',
+    freeze_parameters=['backbone', 'decode_head'],
+    pretrained=
+    'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',
+    backbone=dict(
+        type='MixVisionTransformerCustomInitWeights',
+        in_channels=3,
+        embed_dims=64,
+        num_stages=4,
+        num_layers=[3, 4, 6, 3],
+        num_heads=[1, 2, 5, 8],
+        patch_sizes=[7, 3, 3, 3],
+        sr_ratios=[8, 4, 2, 1],
+        out_indices=(0, 1, 2, 3),
+        mlp_ratio=4,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1),
+    decode_head=dict(
+        type='SegformerHeadUnetFCHeadMultiStepCE',
+        pretrained=
+        'work_dirs2/segformer_mit_b2_segformer_head_unet_fc_single_step_ade_pretrained_freeze_embed_80k_ade20k151/best_mIoU_iter_64000.pth',
+        dim=128,
+        out_dim=256,
+        unet_channels=272,
+        dim_mults=[1, 1, 1],
+        cat_embedding_dim=16,
+        diffusion_timesteps=100,
+        collect_timesteps=[0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 99],
+        in_channels=[64, 128, 320, 512],
+        in_index=[0, 1, 2, 3],
+        channels=256,
+        dropout_ratio=0.1,
+        num_classes=151,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        align_corners=False,
+        ignore_index=0,
+        loss_decode=dict(
+            type='CrossEntropyLoss', use_sigmoid=False, loss_weight=0.1)),
+    train_cfg=dict(),
+    test_cfg=dict(mode='whole'))
+dataset_type = 'ADE20K151Dataset'
+data_root = 'data/ade/ADEChallengeData2016'
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+crop_size = (512, 512)
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', reduce_zero_label=False),
+    dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+    dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+    dict(type='RandomFlip', prob=0.5),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Normalize',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        to_rgb=True),
+    dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(2048, 512),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            dict(type='RandomFlip'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/training',
+        ann_dir='annotations/training',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(type='LoadAnnotations', reduce_zero_label=False),
+            dict(type='Resize', img_scale=(2048, 512), ratio_range=(0.5, 2.0)),
+            dict(type='RandomCrop', crop_size=(512, 512), cat_max_ratio=0.75),
+            dict(type='RandomFlip', prob=0.5),
+            dict(type='PhotoMetricDistortion'),
+            dict(
+                type='Normalize',
+                mean=[123.675, 116.28, 103.53],
+                std=[58.395, 57.12, 57.375],
+                to_rgb=True),
+            dict(type='Pad', size=(512, 512), pad_val=0, seg_pad_val=0),
+            dict(type='DefaultFormatBundle'),
+            dict(type='Collect', keys=['img', 'gt_semantic_seg'])
+        ]),
+    val=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]),
+    test=dict(
+        type='ADE20K151Dataset',
+        data_root='data/ade/ADEChallengeData2016',
+        img_dir='images/validation',
+        ann_dir='annotations/validation',
+        pipeline=[
+            dict(type='LoadImageFromFile'),
+            dict(
+                type='MultiScaleFlipAug',
+                img_scale=(2048, 512),
+                flip=False,
+                transforms=[
+                    dict(type='Resize', keep_ratio=True),
+                    dict(type='RandomFlip'),
+                    dict(
+                        type='Normalize',
+                        mean=[123.675, 116.28, 103.53],
+                        std=[58.395, 57.12, 57.375],
+                        to_rgb=True),
+                    dict(
+                        type='Pad', size_divisor=16, pad_val=0, seg_pad_val=0),
+                    dict(type='ImageToTensor', keys=['img']),
+                    dict(type='Collect', keys=['img'])
+                ])
+        ]))
+log_config = dict(
+    interval=50, hooks=[dict(type='TextLoggerHook', by_epoch=False)])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = None
+resume_from = None
+workflow = [('train', 1)]
+cudnn_benchmark = True
+optimizer = dict(
+    type='AdamW', lr=0.00015, betas=[0.9, 0.96], weight_decay=0.045)
+optimizer_config = dict()
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=1000,
+    warmup_ratio=1e-06,
+    step=20000,
+    gamma=0.5,
+    min_lr=1e-06,
+    by_epoch=False)
+runner = dict(type='IterBasedRunner', max_iters=160000)
+checkpoint_config = dict(by_epoch=False, interval=16000, max_keep_ckpts=1)
+evaluation = dict(
+    interval=16000, metric='mIoU', pre_eval=True, save_best='mIoU')
+custom_hooks = [
+    dict(
+        type='ConstantMomentumEMAHook',
+        momentum=0.01,
+        interval=25,
+        eval_interval=16000,
+        auto_resume=True,
+        priority=49)
+]
+work_dir = './work_dirs2/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce'
+gpu_ids = range(0, 8)
+auto_resume = True

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/best_mIoU_iter_32000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a66cad0c00553fd60ce7f9480e3f7d1df97731fd92aa99695ddc1bf240a6d274
+size 380051503

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/iter_160000.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4c68bd09b726447e49b3a8d30c7a888f069b446c65d971ff6d1eb2e93130120
+size 380051503

ablation/ablation_segformer_mit_b2_segformer_head_unet_fc_small_multi_step_ade_pretrained_freeze_embed_160k_ade20k151_finetune_ema_ce/latest.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4c68bd09b726447e49b3a8d30c7a888f069b446c65d971ff6d1eb2e93130120
+size 380051503