util/misc.py

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# --------------------------------------------------------

import builtins
import datetime
import os
import time
import json
from collections import defaultdict, deque
from pathlib import Path
# from typing import Union

import pandas as pd
import torch
import torch.distributed as dist
import wandb
# from torch._six import inf
from torch import inf
import matplotlib.pyplot as plt
from torchvision import transforms
import cv2
from tqdm import tqdm
from typing import Union, List

class SmoothedValue(object):
    """Track a series of values and provide access to smoothed values over a
    window or the global series average.
    """

    def __init__(self, window_size=20, fmt=None):
        if fmt is None:
            fmt = "{median:.4f} ({global_avg:.4f})"
        self.deque = deque(maxlen=window_size)
        self.total = 0.0
        self.count = 0
        self.fmt = fmt

    def update(self, value, n=1):
        self.deque.append(value)
        self.count += n
        self.total += value * n

    def synchronize_between_processes(self):
        """
        Warning: does not synchronize the deque!
        """
        if not is_dist_avail_and_initialized():
            return
        t = torch.tensor([self.count, self.total], dtype=torch.float64, device='cuda')
        dist.barrier()
        dist.all_reduce(t)
        t = t.tolist()
        self.count = int(t[0])
        self.total = t[1]

    @property
    def median(self):
        d = torch.tensor(list(self.deque))
        return d.median().item()

    @property
    def avg(self):
        d = torch.tensor(list(self.deque), dtype=torch.float32)
        return d.mean().item()

    @property
    def global_avg(self):
        if self.count == 0:
        # Return a default value or handle the zero count scenario
            return 0  # Or any other default value or handling mechanism
        else:
            return self.total / self.count
        # return self.total / self.count

    @property
    def max(self):
        return max(self.deque)

    @property
    def value(self):
        return self.deque[-1]

    def __str__(self):
        return self.fmt.format(
            median=self.median,
            avg=self.avg,
            global_avg=self.global_avg,
            max=self.max,
            value=self.value)


class MetricLogger(object):
    def __init__(self, delimiter="\t"):
        self.meters = defaultdict(SmoothedValue)
        self.delimiter = delimiter

    def update(self, **kwargs):
        for k, v in kwargs.items():
            if v is None:
                continue
            if isinstance(v, torch.Tensor):
                v = v.item()
            assert isinstance(v, (float, int))
            self.meters[k].update(v)

    def __getattr__(self, attr):
        if attr in self.meters:
            return self.meters[attr]
        if attr in self.__dict__:
            return self.__dict__[attr]
        raise AttributeError("'{}' object has no attribute '{}'".format(
            type(self).__name__, attr))

    def __str__(self):
        loss_str = []
        for name, meter in self.meters.items():
            loss_str.append(
                "{}: {}".format(name, str(meter))
            )
        return self.delimiter.join(loss_str)

    def synchronize_between_processes(self):
        for meter in self.meters.values():
            meter.synchronize_between_processes()

    def add_meter(self, name, meter):
        self.meters[name] = meter

    def log_every(self, iterable, print_freq, header=None):
        i = 0
        if not header:
            header = ''
        start_time = time.time()
        end = time.time()
        iter_time = SmoothedValue(fmt='{avg:.4f}')
        data_time = SmoothedValue(fmt='{avg:.4f}')
        space_fmt = ':' + str(len(str(len(iterable)))) + 'd'
        log_msg = [
            header,
            '[{0' + space_fmt + '}/{1}]',
            'eta: {eta}',
            '{meters}',
            'time: {time}',
            'data: {data}'
        ]
        if torch.cuda.is_available():
            log_msg.append('max mem: {memory:.0f}')
        log_msg = self.delimiter.join(log_msg)
        MB = 1024.0 * 1024.0
        for obj in iterable:
            data_time.update(time.time() - end)
            yield obj
            iter_time.update(time.time() - end)
            if i % print_freq == 0 or i == len(iterable) - 1:
                eta_seconds = iter_time.global_avg * (len(iterable) - i)
                eta_string = str(datetime.timedelta(seconds=int(eta_seconds)))
                if torch.cuda.is_available():
                    print(log_msg.format(
                        i, len(iterable), eta=eta_string,
                        meters=str(self),
                        time=str(iter_time), data=str(data_time),
                        memory=torch.cuda.max_memory_allocated() / MB))
                else:
                    print(log_msg.format(
                        i, len(iterable), eta=eta_string,
                        meters=str(self),
                        time=str(iter_time), data=str(data_time)))
            i += 1
            end = time.time()
        total_time = time.time() - start_time
        total_time_str = str(datetime.timedelta(seconds=int(total_time)))
        print('{} Total time: {} ({:.4f} s / it)'.format(
            header, total_time_str, total_time / len(iterable)))


def setup_for_distributed(is_master):
    """
    This function disables printing when not in master process
    """
    builtin_print = builtins.print

    def print(*args, **kwargs):
        force = kwargs.pop('force', False)
        force = force or (get_world_size() > 8)
        if is_master or force:
            now = datetime.datetime.now().time()
            builtin_print('[{}] '.format(now), end='')  # print with time stamp
            builtin_print(*args, **kwargs)

    builtins.print = print


def is_dist_avail_and_initialized():
    if not dist.is_available():
        return False
    if not dist.is_initialized():
        return False
    return True


def get_world_size():
    if not is_dist_avail_and_initialized():
        return 1
    return dist.get_world_size()


def get_rank():
    if not is_dist_avail_and_initialized():
        return 0
    return dist.get_rank()


def is_main_process():
    return get_rank() == 0


def save_on_master(*args, **kwargs):
    if is_main_process():
        torch.save(*args, **kwargs)


def init_distributed_mode(args):
    if args.dist_on_itp:
        args.rank = int(os.environ['OMPI_COMM_WORLD_RANK'])
        args.world_size = int(os.environ['OMPI_COMM_WORLD_SIZE'])
        args.gpu = int(os.environ['OMPI_COMM_WORLD_LOCAL_RANK'])
        args.dist_url = "tcp://%s:%s" % (os.environ['MASTER_ADDR'], os.environ['MASTER_PORT'])
        os.environ['LOCAL_RANK'] = str(args.gpu)
        os.environ['RANK'] = str(args.rank)
        os.environ['WORLD_SIZE'] = str(args.world_size)
        # ["RANK", "WORLD_SIZE", "MASTER_ADDR", "MASTER_PORT", "LOCAL_RANK"]
    elif 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
        args.rank = int(os.environ["RANK"])
        args.world_size = int(os.environ['WORLD_SIZE'])
        args.gpu = int(os.environ['LOCAL_RANK'])
    elif 'SLURM_PROCID' in os.environ:
        args.rank = int(os.environ['SLURM_PROCID'])
        args.gpu = args.rank % torch.cuda.device_count()
    else:
        print('Not using distributed mode')
        setup_for_distributed(is_master=True)  # hack
        args.distributed = False
        return

    args.distributed = True

    torch.cuda.set_device(args.gpu)
    args.dist_backend = 'nccl'
    print('| distributed init (rank {}): {}, gpu {}'.format(
        args.rank, args.dist_url, args.gpu), flush=True)
    torch.distributed.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
                                         world_size=args.world_size, rank=args.rank)
    torch.distributed.barrier()
    setup_for_distributed(args.rank == 0)


class NativeScalerWithGradNormCount:
    state_dict_key = "amp_scaler"

    def __init__(self):
        self._scaler = torch.cuda.amp.GradScaler()

    def __call__(self, loss, optimizer, clip_grad=None, parameters=None, create_graph=False, update_grad=True):
        self._scaler.scale(loss).backward(create_graph=create_graph)
        if update_grad:
            if clip_grad is not None:
                assert parameters is not None
                self._scaler.unscale_(optimizer)  # unscale the gradients of optimizer's assigned params in-place
                norm = torch.nn.utils.clip_grad_norm_(parameters, clip_grad)
            else:
                self._scaler.unscale_(optimizer)
                norm = get_grad_norm_(parameters)
            self._scaler.step(optimizer)
            self._scaler.update()
        else:
            norm = None
        return norm

    def state_dict(self):
        return self._scaler.state_dict()

    def load_state_dict(self, state_dict):
        self._scaler.load_state_dict(state_dict)


def get_grad_norm_(parameters, norm_type: float = 2.0) -> torch.Tensor:
    if isinstance(parameters, torch.Tensor):
        parameters = [parameters]
    parameters = [p for p in parameters if p.grad is not None]
    norm_type = float(norm_type)
    if len(parameters) == 0:
        return torch.tensor(0.)
    device = parameters[0].grad.device
    if norm_type == inf:
        total_norm = max(p.grad.detach().abs().max().to(device) for p in parameters)
    else:
        total_norm = torch.norm(torch.stack([torch.norm(p.grad.detach(), norm_type).to(device) for p in parameters]), norm_type)
    return total_norm


def save_model(args, epoch, model, model_without_ddp, optimizer, loss_scaler, suffix="", upload=True):
    if suffix:
        suffix = f"__{suffix}"
    output_dir = Path(args.output_dir)
    ckpt_name = f"checkpoint{suffix}.pth"
    if loss_scaler is not None:
        checkpoint_paths = [output_dir / ckpt_name]
        for checkpoint_path in checkpoint_paths:
            to_save = {
                'model': model_without_ddp.state_dict(),
                'optimizer': optimizer.state_dict(),
                'epoch': epoch,
                'scaler': loss_scaler.state_dict(),
                'args': args,
            }
            save_on_master(to_save, checkpoint_path)
            if upload and is_main_process():
                log_wandb_model(f"checkpoint{suffix}", checkpoint_path, epoch)
            print("checkpoint sent to W&B (if)")
    else:
        client_state = {'epoch': epoch}
        model.save_checkpoint(save_dir=args.output_dir, tag=ckpt_name, client_state=client_state)
        if upload and is_main_process():
            log_wandb_model(f"checkpoint{suffix}", output_dir / ckpt_name, epoch)
        print("checkpoint sent to W&B (else)")


def log_wandb_model(title, path, epoch):
    artifact = wandb.Artifact(title, type="model")
    artifact.add_file(path)
    artifact.metadata["epoch"] = epoch
    wandb.log_artifact(artifact_or_path=artifact, name=title)


def load_model(args, model_without_ddp, optimizer, loss_scaler):
    if args.resume:
        if args.resume.startswith('https'):
            checkpoint = torch.hub.load_state_dict_from_url(
                args.resume, map_location='cpu', check_hash=True)
        else:
            checkpoint = torch.load(args.resume, map_location='cpu')

        if 'pos_embed' in checkpoint['model'] and checkpoint['model']['pos_embed'].shape != model_without_ddp.state_dict()['pos_embed'].shape:
            print(f"Removing key pos_embed from pretrained checkpoint")
            del checkpoint['model']['pos_embed']

        if 'decoder_pos_embed' in checkpoint['model'] and checkpoint['model']['decoder_pos_embed'].shape != model_without_ddp.state_dict()['decoder_pos_embed'].shape:
            print(f"Removing key decoder_pos_embed from pretrained checkpoint")
            del checkpoint['model']['decoder_pos_embed']

        model_without_ddp.load_state_dict(checkpoint['model'], strict=False)
        print("Resume checkpoint %s" % args.resume)
        if 'optimizer' in checkpoint and 'epoch' in checkpoint and not (hasattr(args, 'eval') and args.eval):
            optimizer.load_state_dict(checkpoint['optimizer'])
            args.start_epoch = checkpoint['epoch'] + 1
            if 'scaler' in checkpoint:
                loss_scaler.load_state_dict(checkpoint['scaler'])
            print("With optim & sched!")

def load_model_FSC(args, model_without_ddp):
    if args.resume:
        if args.resume.startswith('https'):
            checkpoint = torch.hub.load_state_dict_from_url(
                args.resume, map_location='cpu', check_hash=True)
        else:
            checkpoint = torch.load(args.resume, map_location='cpu')

        if 'pos_embed' in checkpoint['model'] and checkpoint['model']['pos_embed'].shape != model_without_ddp.state_dict()['pos_embed'].shape:
            print(f"Removing key pos_embed from pretrained checkpoint")
            del checkpoint['model']['pos_embed']

        model_without_ddp.load_state_dict(checkpoint['model'], strict=False)
        print(f"Resume checkpoint {args.resume} ({checkpoint['epoch']})")

def load_model_FSC1(args, model_without_ddp):
    if args.resume:
        if args.resume.startswith('https'):
            checkpoint = torch.hub.load_state_dict_from_url(
                args.resume, map_location='cpu', check_hash=True)
        else:
            checkpoint = torch.load(args.resume, map_location='cpu')
            #model = timm.create_model('vit_base_patch16_224', pretrained=True)
            #torch.save(model.state_dict(), './output_abnopre_dir/checkpoint-6657.pth')
            checkpoint1 = torch.load('./output_abnopre_dir/checkpoint-6657.pth', map_location='cpu')

        if 'pos_embed' in checkpoint['model'] and checkpoint['model']['pos_embed'].shape != model_without_ddp.state_dict()['pos_embed'].shape:
            print(f"Removing key pos_embed from pretrained checkpoint")
            del checkpoint['model']['pos_embed']

        del checkpoint1['cls_token'],checkpoint1['pos_embed']

        model_without_ddp.load_state_dict(checkpoint['model'], strict=False)
        model_without_ddp.load_state_dict(checkpoint1, strict=False)
        print("Resume checkpoint %s" % args.resume)


def load_model_FSC_full(args, model_without_ddp, optimizer, loss_scaler):
    if args.resume:
        if args.resume.startswith('https'):
            checkpoint = torch.hub.load_state_dict_from_url(
                args.resume, map_location='cpu', check_hash=True)
        else:
            checkpoint = torch.load(args.resume, map_location='cpu')

        if 'pos_embed' in checkpoint['model'] and checkpoint['model']['pos_embed'].shape != \
                model_without_ddp.state_dict()['pos_embed'].shape:
            print(f"Removing key pos_embed from pretrained checkpoint")
            del checkpoint['model']['pos_embed']

        model_without_ddp.load_state_dict(checkpoint['model'], strict=False)
        print("Resume checkpoint %s" % args.resume)

        if 'optimizer' in checkpoint and 'epoch' in checkpoint and args.do_resume:
            optimizer.load_state_dict(checkpoint['optimizer'])
            args.start_epoch = checkpoint['epoch'] + 1
            if 'scaler' in checkpoint:
                loss_scaler.load_state_dict(checkpoint['scaler'])
            print("With optim & scheduler!")


def all_reduce_mean(x):
    world_size = get_world_size()
    if world_size > 1:
        x_reduce = torch.tensor(x).cuda()
        dist.all_reduce(x_reduce)
        x_reduce /= world_size
        return x_reduce.item()
    else:
        return x


def plot_counts(res_csv: Union[str, List[str]], output_dir: str, suffix: str = "", smooth: bool = False):
    if suffix:
        suffix = f"_{suffix}"
    if smooth:
        suffix = f"_smooth{suffix}"
    if type(res_csv) == str:
        res_csv = [res_csv]

    plt.figure(figsize=(15, 5))

    for res in res_csv:
        name = Path(res).parent.name
        df = pd.read_csv(res)
        print(df)

        df.sort_values(by="name", inplace=True)
        df.reset_index(drop=True, inplace=True)
        df.index += 1
        print(df)

        if smooth:
            time_arr = df.index[5:-5]
            smooth_pred_mean = df['prediction'].iloc[5:-5].rolling(25).mean()
            smooth_pred_std = df['prediction'].iloc[5:-5].rolling(25).std()
            plt.plot(time_arr, smooth_pred_mean, label=name)
            plt.fill_between(time_arr, smooth_pred_mean + smooth_pred_std, smooth_pred_mean - smooth_pred_std, alpha=.2)
            plt.xlabel('Frame')
            plt.ylabel('Count')
        else:
            plt.plot(df.index, df['prediction'], label=name)

    plt.legend()
    plt.savefig(os.path.join(output_dir, f'counts{suffix}.png'), dpi=300)


def write_zeroshot_annotations(p: Path):
    with open(p / 'annotations.json', 'a') as split:
        split.write('{\n')
        for img in p.iterdir():
            if img.is_file():
                split.write(f'  "{img.name}": {{\n' \
                            '    "H": 960,\n' \
                            '    "W": 1280,\n' \
                            '    "box_examples_coordinates": [],\n' \
                            '    "points": []\n' \
                            '  },\n')
        split.write("}")

    with open(p / 'split.json', 'a') as split:
        split.write('{\n  "test":\n  [\n')
        for img in p.iterdir():
            if img.is_file():
                split.write(f'    "{img.name}",\n')
        split.write("  ]\n}")


def make_grid(imgs, h, w):
    assert len(imgs) == 9
    rows = []
    for i in range(0, 9, 3):
        row = torch.cat((imgs[i], imgs[i + 1], imgs[i + 2]), -1)
        rows += [row]
    grid = torch.cat((rows[0], rows[1], rows[2]), 0)
    grid = transforms.Resize((h, w))(grid.unsqueeze(0))
    return grid.squeeze(0)


def min_max(t):
    t_shape = t.shape
    t = t.view(t_shape[0], -1)
    t -= t.min(1, keepdim=True)[0]
    t /= t.max(1, keepdim=True)[0]
    t = t.view(*t_shape)
    return t


def min_max_np(v, new_min=0, new_max=1):
    v_min, v_max = v.min(), v.max()
    return (v - v_min) / (v_max - v_min) * (new_max - new_min) + new_min


def get_box_map(sample, pos, device, external=False):
    box_map = torch.zeros([sample.shape[1], sample.shape[2]], device=device)
    if external is False:
        for rect in pos:
            for i in range(rect[2] - rect[0]):
                box_map[min(rect[0] + i, sample.shape[1] - 1), min(rect[1], sample.shape[2] - 1)] = 10
                box_map[min(rect[0] + i, sample.shape[1] - 1), min(rect[3], sample.shape[2] - 1)] = 10
            for i in range(rect[3] - rect[1]):
                box_map[min(rect[0], sample.shape[1] - 1), min(rect[1] + i, sample.shape[2] - 1)] = 10
                box_map[min(rect[2], sample.shape[1] - 1), min(rect[1] + i, sample.shape[2] - 1)] = 10
        box_map = box_map.unsqueeze(0).repeat(3, 1, 1)
    return box_map


timerfunc = time.perf_counter

class measure_time(object):
    def __enter__(self):
        self.start = timerfunc()
        return self

    def __exit__(self, typ, value, traceback):
        self.duration = timerfunc() - self.start

    def __add__(self, other):
        return self.duration + other.duration

    def __sub__(self, other):
        return self.duration - other.duration
    
    def __str__(self):
        return str(self.duration)


def log_test_results(test_dir):
    test_dir = Path(test_dir)
    logs = []
    for d in test_dir.iterdir():
        if d.is_dir() and (d / "log.txt").exists():
            print(d.name)
            with open(d / "log.txt") as f:
                last = f.readlines()[-1]
                j = json.loads(last)
                j['name'] = d.name
                logs.append(j)
    df = pd.DataFrame(logs)

    df.sort_values('name', inplace=True, ignore_index=True)
    cols = list(df.columns)
    cols = cols[-1:] + cols[:-1]
    df = df[cols]

    df.to_csv(test_dir / "logs.csv", index=False)


COLORS = {
    'muted blue': '#1f77b4',
    'safety orange': '#ff7f0e',
    'cooked asparagus green': '#2ca02c',
    'brick red': '#d62728',
    'muted purple': '#9467bd',
    'chestnut brown': '#8c564b',
    'raspberry yogurt pink': '#e377c2',
    'middle gray': '#7f7f7f',
    'curry yellow-green': '#bcbd22',
    'blue-teal': '#17becf',
    'muted blue light': '#419ede',
    'safety orange light': '#ffa85b',
    'cooked asparagus green light': '#4bce4b',
    'brick red light': '#e36667'
}


def plot_test_results(test_dir):
    import plotly.graph_objects as go

    test_dir = Path(test_dir)
    df = pd.read_csv(test_dir / "logs.csv")
    df.sort_values('name', inplace=True)

    fig = go.Figure()
    fig.add_trace(go.Scatter(x=df['name'], y=df['MAE'], line_color=COLORS['muted blue'],
                        mode='lines', name='MAE'))
    fig.add_trace(go.Scatter(x=df['name'], y=df['RMSE'], line_color=COLORS['safety orange'],
                        mode='lines', name='RMSE'))
    fig.add_trace(go.Scatter(x=df['name'], y=df['NAE'], line_color=COLORS['cooked asparagus green'],
                        mode='lines', name='NAE'))

    fig.update_yaxes(type="log")
    fig.write_image(test_dir / "plot.jpeg", scale=4)
    fig.write_html(test_dir / "plot.html", auto_open=False)


def frames2vid(input_dir: str, output_file: str, pattern: str, fps: int, h=720, w=1280):
    input_dir = Path(input_dir)
    video_file = None
    files = sorted(input_dir.glob(pattern))
    video_file = cv2.VideoWriter(output_file, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
    for img in tqdm(files, total=len(files)):
        frame = cv2.imread(str(img))
        frame = cv2.resize(frame, (w, h))
        video_file.write(frame)

    video_file.release()