Initial commit

README.md

@@ -42,7 +42,7 @@ with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
         ...
 ```
 
-Refer to the [demo notebooks](https://github.com/facebookresearch/segment-anything-2/tree/main/notebooks) for details.
+Refer to the [demo notebooks](https://github.com/facebookresearch/segment-anything-2/tree/main/notebooks) for more details.
 
 ### Citation
 
@@ -55,4 +55,4 @@ To cite the paper, model, or software, please use the below:
   url={https://arxiv.org/abs/2408.00714},
   year={2024}
 }
-```
+```

sam2_hiera_l.yaml → sam2.1_hiera_l.yaml

@@ -93,6 +93,7 @@ model:
   use_mask_input_as_output_without_sam: true
   # Memory
   directly_add_no_mem_embed: true
+  no_obj_embed_spatial: true
   # use high-resolution feature map in the SAM mask decoder
   use_high_res_features_in_sam: true
   # output 3 masks on the first click on initial conditioning frames
@@ -101,7 +102,9 @@ model:
   iou_prediction_use_sigmoid: True
   # cross-attend to object pointers from other frames (based on SAM output tokens) in the encoder
   use_obj_ptrs_in_encoder: true
-  add_tpos_enc_to_obj_ptrs: false
+  add_tpos_enc_to_obj_ptrs: true
+  proj_tpos_enc_in_obj_ptrs: true
+  use_signed_tpos_enc_to_obj_ptrs: true
   only_obj_ptrs_in_the_past_for_eval: true
   # object occlusion prediction
   pred_obj_scores: true
@@ -114,4 +117,4 @@ model:
   multimask_max_pt_num: 1
   use_mlp_for_obj_ptr_proj: true
   # Compilation flag
-  compile_image_encoder: False
+  compile_image_encoder: False

sam2.1_hiera_large.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b36b71d5cafc83a0975d14d0afae81c3915804e12cc896b0665eaabcc445d56
-size 898089053
+oid sha256:2647878d5dfa5098f2f8649825738a9345572bae2d4350a2468587ece47dd318
+size 898083611