utils.py

import io
import os
import math
import time
import json
from collections import defaultdict, deque
import datetime
import numpy as np
import random
from timm.utils import get_state_dict
from torch.utils.data._utils.collate import default_collate
from pathlib import Path
import subprocess
import torch
import torch.distributed as dist
#from torch import inf
from torch import inf

from PIL import Image
from visualize_masks import denormalize_video

from tensorboardX import SummaryWriter
# import faiss

import torch.nn.functional as F


def _no_grad_trunc_normal_(tensor, mean, std, a, b):
    # Cut & paste from PyTorch official master until it's in a few official releases - RW
    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
    def norm_cdf(x):
        # Computes standard normal cumulative distribution function
        return (1. + math.erf(x / math.sqrt(2.))) / 2.

    if (mean < a - 2 * std) or (mean > b + 2 * std):
        warnings.warn("mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
                      "The distribution of values may be incorrect.",
                      stacklevel=2)

    with torch.no_grad():
        # Values are generated by using a truncated uniform distribution and
        # then using the inverse CDF for the normal distribution.
        # Get upper and lower cdf values
        l = norm_cdf((a - mean) / std)
        u = norm_cdf((b - mean) / std)

        # Uniformly fill tensor with values from [l, u], then translate to
        # [2l-1, 2u-1].
        tensor.uniform_(2 * l - 1, 2 * u - 1)

        # Use inverse cdf transform for normal distribution to get truncated
        # standard normal
        tensor.erfinv_()

        # Transform to proper mean, std
        tensor.mul_(std * math.sqrt(2.))
        tensor.add_(mean)

        # Clamp to ensure it's in the proper range
        tensor.clamp_(min=a, max=b)
        return tensor


def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
    # type: (Tensor, float, float, float, float) -> Tensor
    return _no_grad_trunc_normal_(tensor, mean, std, a, b)



def mask_fix(feat, mask):
    B, D, C = feat.shape
    
    #mask features
    mask_feat = []
    for f, m in zip(feat, mask):
        mask_feat.append(f[m])
    
    # get longest dimensional feature
    max_dim = 0
    for f in mask_feat:
        dim = len(f)
        if dim > max_dim:
            max_dim = dim
            
    #append 0 to match dimensions
    out_feat = []
    for f in mask_feat:
        diff = max_dim - len(f)
        if diff > 0:
            f = torch.cat((f, torch.zeros((diff, C)).to(feat.device)), dim=0)
        out_feat.append(f.unsqueeze(0))
    
    mask_feat = torch.cat((out_feat), dim=0)
    assert len(mask_feat.shape) == 3
    return mask_feat

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):
        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)))


class TensorboardLogger(object):
    def __init__(self, log_dir):
        self.writer = SummaryWriter(logdir=log_dir)
        self.step = 0

    def set_step(self, step=None):
        if step is not None:
            self.step = step
        else:
            self.step += 1

    def update(self, head='scalar', step=None, **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.writer.add_scalar(head + "/" + k, v, self.step if step is None else step)

    def flush(self):
        self.writer.flush()

def seed_worker(worker_id):
    worker_seed = torch.initial_seed() % 2**32
    np.random.seed(worker_seed)
    random.seed(worker_seed)
    
def _load_checkpoint_for_ema(model_ema, checkpoint):
    """
    Workaround for ModelEma._load_checkpoint to accept an already-loaded object
    """
    mem_file = io.BytesIO()
    torch.save(checkpoint, mem_file)
    mem_file.seek(0)
    model_ema._load_checkpoint(mem_file)


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

    def print(*args, **kwargs):
        force = kwargs.pop('force', False)
        if is_master or force:
            builtin_print(*args, **kwargs)

    __builtin__.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)
    #elif 'SLURM_PROCID' in os.environ:
    #    args.rank = int(os.environ['SLURM_PROCID'])
    #    args.gpu = int(os.environ['SLURM_LOCALID'])
    #    args.world_size = int(os.environ['SLURM_NTASKS'])
    #    os.environ['RANK'] = str(args.rank)
    #    os.environ['LOCAL_RANK'] = str(args.gpu)
    #    os.environ['WORLD_SIZE'] = str(args.world_size)

    #    node_list = os.environ['SLURM_NODELIST']
    #    addr = subprocess.getoutput(
    #        f'scontrol show hostname {node_list} | head -n1')
    #    if 'MASTER_ADDR' not in os.environ:
    #        os.environ['MASTER_ADDR'] = addr
    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'])
    else:
        print('Not using distributed mode')
        args.distributed = False
        return

    args.distributed = True
    print(f"gpu {args.gpu}, world_size {args.world_size}, rank {args.rank}")
    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()
    # assert torch.distributed.is_initialized()
    setup_for_distributed(args.rank == 0)


def load_state_dict(model, state_dict, prefix='', ignore_missing="relative_position_index"):
    missing_keys = []
    unexpected_keys = []
    error_msgs = []
    metadata = getattr(state_dict, '_metadata', None)
    state_dict = state_dict.copy()
    if metadata is not None:
        state_dict._metadata = metadata

    def load(module, prefix=''):
        local_metadata = {} if metadata is None else metadata.get(
            prefix[:-1], {})
        module._load_from_state_dict(
            state_dict, prefix, local_metadata, True, missing_keys, unexpected_keys, error_msgs)
        for name, child in module._modules.items():
            if child is not None:
                load(child, prefix + name + '.')

    load(model, prefix=prefix)

    warn_missing_keys = []
    ignore_missing_keys = []
    for key in missing_keys:
        keep_flag = True
        for ignore_key in ignore_missing.split('|'):
            if ignore_key in key:
                keep_flag = False
                break
        if keep_flag:
            warn_missing_keys.append(key)
        else:
            ignore_missing_keys.append(key)

    missing_keys = warn_missing_keys

    if len(missing_keys) > 0:
        print("Weights of {} not initialized from pretrained model: {}".format(
            model.__class__.__name__, missing_keys))
    if len(unexpected_keys) > 0:
        print("Weights from pretrained model not used in {}: {}".format(
            model.__class__.__name__, unexpected_keys))
    if len(ignore_missing_keys) > 0:
        print("Ignored weights of {} not initialized from pretrained model: {}".format(
            model.__class__.__name__, ignore_missing_keys))
    if len(error_msgs) > 0:
        print('\n'.join(error_msgs))


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 cosine_scheduler(base_value, final_value, epochs, niter_per_ep, warmup_epochs=0,
                     start_warmup_value=0, warmup_steps=-1):
    warmup_schedule = np.array([])
    warmup_iters = warmup_epochs * niter_per_ep
    if warmup_steps > 0:
        warmup_iters = warmup_steps
    print("Set warmup steps = %d" % warmup_iters)
    if warmup_epochs > 0:
        warmup_schedule = np.linspace(start_warmup_value, base_value, warmup_iters)

    iters = np.arange(epochs * niter_per_ep - warmup_iters)
    schedule = np.array(
        [final_value + 0.5 * (base_value - final_value) * (1 + math.cos(math.pi * i / (len(iters)))) for i in iters])

    schedule = np.concatenate((warmup_schedule, schedule))

    assert len(schedule) == epochs * niter_per_ep
    return schedule


def save_model(args, epoch, model, model_without_ddp, optimizer, loss_scaler, model_ema=None):
    output_dir = Path(args.output_dir)
    epoch_name = str(epoch)
    if loss_scaler is not None:
        checkpoint_paths = [output_dir / ('checkpoint-%s.pth' % epoch_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,
            }

            if model_ema is not None:
                to_save['model_ema'] = get_state_dict(model_ema)

            save_on_master(to_save, checkpoint_path)
    else:
        client_state = {'epoch': epoch}
        if model_ema is not None:
            client_state['model_ema'] = get_state_dict(model_ema)
        model.save_checkpoint(save_dir=args.output_dir, tag="checkpoint-%s" % epoch_name, client_state=client_state)

def auto_load_best_model(args, model, model_without_ddp, optimizer, loss_scaler, model_ema=None):
    output_dir = Path(args.output_dir)
    if loss_scaler is not None:
        # torch.amp
        if args.auto_resume and len(args.resume) == 0:
            import glob
            all_checkpoints = glob.glob(os.path.join(output_dir, 'checkpoint-*.pth'))
            latest_ckpt = -1
            for ckpt in all_checkpoints:
                t = ckpt.split('-')[-1].split('.')[0]
                if t.isdigit():
                    latest_ckpt = max(int(t), latest_ckpt)
            if latest_ckpt >= 0:
                args.resume = os.path.join(output_dir, 'checkpoint-%d.pth' % latest_ckpt)

            if args.eval:
                args.resume = os.path.join(output_dir, 'checkpoint-best.pth')
            print("Auto resume checkpoint: %s" % args.resume)

        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_without_ddp.load_state_dict(checkpoint['model'])
            print("Resume checkpoint %s" % args.resume)


def auto_load_model(args, model, model_without_ddp, optimizer, loss_scaler, model_ema=None):
    output_dir = Path(args.output_dir)
    if loss_scaler is not None:
        # torch.amp
        if args.auto_resume and len(args.resume) == 0:
            import glob
            all_checkpoints = glob.glob(os.path.join(output_dir, 'checkpoint-*.pth'))
            latest_ckpt = -1
            for ckpt in all_checkpoints:
                t = ckpt.split('-')[-1].split('.')[0]
                if t.isdigit():
                    latest_ckpt = max(int(t), latest_ckpt)
            if latest_ckpt >= 0:
                args.resume = os.path.join(output_dir, 'checkpoint-%d.pth' % latest_ckpt)

            print("Auto resume checkpoint: %s" % args.resume)

        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_without_ddp.load_state_dict(checkpoint['model'])
            print("Resume checkpoint %s" % args.resume)
            if 'optimizer' in checkpoint and 'epoch' in checkpoint:
                optimizer.load_state_dict(checkpoint['optimizer'])
                args.start_epoch = checkpoint['epoch'] + 1
                if hasattr(args, 'model_ema') and args.model_ema:
                    _load_checkpoint_for_ema(model_ema, checkpoint['model_ema'])
                if 'scaler' in checkpoint:
                    loss_scaler.load_state_dict(checkpoint['scaler'])
                print("With optim & sched!")
    else:
        # deepspeed, only support '--auto_resume'.
        if args.auto_resume:
            import glob
            all_checkpoints = glob.glob(os.path.join(output_dir, 'checkpoint-*'))
            latest_ckpt = -1
            for ckpt in all_checkpoints:
                t = ckpt.split('-')[-1].split('.')[0]
                if t.isdigit():
                    latest_ckpt = max(int(t), latest_ckpt)
            if latest_ckpt >= 0:
                args.resume = os.path.join(output_dir, 'checkpoint-%d' % latest_ckpt)
                print("Auto resume checkpoint: %d" % latest_ckpt)
                _, client_states = model.load_checkpoint(args.output_dir, tag='checkpoint-%d' % latest_ckpt)
                args.start_epoch = client_states['epoch'] + 1
                if model_ema is not None:
                    if args.model_ema:
                        _load_checkpoint_for_ema(model_ema, client_states['model_ema'])


def create_ds_config(args):
    args.deepspeed_config = os.path.join(args.output_dir, "deepspeed_config.json")
    with open(args.deepspeed_config, mode="w") as writer:
        ds_config = {
            "train_batch_size": args.batch_size * args.update_freq * get_world_size(),
            "train_micro_batch_size_per_gpu": args.batch_size,
            "steps_per_print": 1000,
            "optimizer": {
                "type": "Adam",
                "adam_w_mode": True,
                "params": {
                    "lr": args.lr,
                    "weight_decay": args.weight_decay,
                    "bias_correction": True,
                    "betas": [
                        0.9,
                        0.999
                    ],
                    "eps": 1e-8
                }
            },
            "fp16": {
                "enabled": True,
                "loss_scale": 0,
                "initial_scale_power": 7,
                "loss_scale_window": 128
            }
        }

        writer.write(json.dumps(ds_config, indent=2))

def multiple_samples_collate(batch, fold=False):
    """
    Collate function for repeated augmentation. Each instance in the batch has
    more than one sample.
    Args:
        batch (tuple or list): data batch to collate.
    Returns:
        (tuple): collated data batch.
    """
    inputs, labels, video_idx, extra_data = zip(*batch)
    inputs = [item for sublist in inputs for item in sublist]
    labels = [item for sublist in labels for item in sublist]
    video_idx = [item for sublist in video_idx for item in sublist]
    inputs, labels, video_idx, extra_data = (
        default_collate(inputs),
        default_collate(labels),
        default_collate(video_idx),
        default_collate(extra_data),
    )
    if fold:
        return [inputs], labels, video_idx, extra_data
    else:
        return inputs, labels, video_idx, extra_data


def cluster_features(features, k):
    """
    Cluster features into k clusters using k-means clustering.
    Args:
        features (torch.Tensor): features to cluster [T, np, dim].
        k (int): number of clusters.
    Returns:
        (torch.Tensor): cluster map in which each patch is assigned to a cluster in one-hot.
    """
    kmeans = faiss.Kmeans(features.size(-1), k, niter=20, verbose=True, gpu=True)
    T, np, dim = features.shape
    features = features.reshape(T * np, dim)
    feature_numpy = features.cpu().detach().numpy()
    kmeans.train(feature_numpy)
    D, I = kmeans.index.search(feature_numpy, 1)
    cluster_maps = torch.from_numpy(I.reshape(T, np))
    one_hot_cluster_maps = F.one_hot(cluster_maps, k)
    return one_hot_cluster_maps



def random_masking(x, mask_ratio):
    """
    Perform per-sample random masking by per-sample shuffling.
    Per-sample shuffling is done by argsort random noise.
    x: [N, L, D], sequence
    """
    N, L, D = x.shape  # batch, length, dim
    len_keep = int(L * (1 - mask_ratio))

    noise = torch.rand(N, L, device=x.device)  # noise in [0, 1]

    # sort noise for each sample
    ids_shuffle = torch.argsort(noise, dim=1)  # ascend: small is keep, large is remove
    ids_restore = torch.argsort(ids_shuffle, dim=1)

    # keep the first subset
    ids_keep = ids_shuffle[:, :len_keep]
    # remove the second subset
    ids_remove = ids_shuffle[:, len_keep:]

    # generate the binary mask: 0 is keep, 1 is remove
    mask = torch.ones([N, L], device=x.device)
    mask[:, :len_keep] = 0
    # unshuffle to get the binary mask
    mask = torch.gather(mask, dim=1, index=ids_restore).bool()

    return ids_keep, mask, ids_restore, ids_remove


@torch.no_grad()
def sinkhorn(Q: torch.Tensor, nmb_iters: int, world_size=1) -> torch.Tensor:
    with torch.no_grad():
        Q = Q.detach().clone()
        sum_Q = torch.sum(Q)
        if world_size > 1:
            dist.all_reduce(sum_Q)
        Q /= sum_Q
        K, B = Q.shape
        u = torch.zeros(K).to(Q.device)
        r = torch.ones(K).to(Q.device) / K
        c = torch.ones(B).to(Q.device) / B * world_size

        if world_size > 1:
            curr_sum = torch.sum(Q, dim=1)
            dist.all_reduce(curr_sum)

        for _ in range(nmb_iters):
            if world_size > 1:
                u = curr_sum
            else:
                u = torch.sum(Q, dim=1)
            Q *= (r / u).unsqueeze(1)
            Q *= (c / torch.sum(Q, dim=0)).unsqueeze(0)
            if world_size > 1:
                curr_sum = torch.sum(Q, dim=1)
                dist.all_reduce(curr_sum)

        return (Q / torch.sum(Q, dim=0, keepdim=True)).t().float()


def chromatic_correction_fct(tensor):
    # Define the green-magenta color axis in RGB space
    a = torch.tensor([-1, 2, -1], dtype=torch.float32).view(3, 1)
    
    # Compute the outer product of a with itself and the dot product
    outer_product = torch.mm(a, a.t())
    dot_product = torch.mm(a.t(), a)
    
    # Compute the projection matrix B
    B = torch.eye(3) - outer_product / dot_product
    
    # Apply the matrix B to every pixel in the tensor
    # Reshape the tensor to apply the transformation and then revert to original shape
    transformed_tensor = tensor.permute(0, 2, 3, 4, 1)
    original_shape = transformed_tensor.shape
    transformed_tensor = torch.matmul(transformed_tensor.reshape(-1, 3), B.to(tensor.device)).reshape(original_shape)
    transformed_tensor = transformed_tensor.permute(0, 4, 1, 2, 3)
    # visualization purpose
    # step =  random.randint(0, 1000)
    # save_processed_video(transformed_tensor, "/home/michael/runs/videos/corr_", step)
    # save_processed_video(tensor, "/home/michael/runs/videos/true_", step)
    return transformed_tensor


def rescale_pixel_values(frame):
    """
    Rescales the pixel values of the image to span the full 0-255 range.
    
    Parameters:
    - frame: A numpy array representing the image with pixel values.
    
    Returns:
    - A new numpy array with pixel values rescaled to 0-255.
    """
    # Find the current minimum and maximum values of the frame
    min_val = frame.min()
    max_val = frame.max()
    
    # Perform the rescaling to the range 0-255
    rescaled_frame = (frame - min_val) / (max_val - min_val) * 255
    
    # Convert to an unsigned 8-bit integer type to get final 0-255 range values
    rescaled_frame = rescaled_frame.astype(np.uint8)
    
    return rescaled_frame

def save_processed_video(unnorm_videos, output_dir, step):
    """
    Process and save a video with an overlay to a specified directory.

    Parameters:
    - unnorm_videos: Tensor, the video data to be processed.
    - output_dir: str, the directory where the processed video will be saved.
    - step: int, used for naming the output file.
    - denormalize_video: function, used to denormalize the video data.
    - overlay_video_cmap: function, used to overlay the clustering map on the video frames.
    """
    videos = unnorm_videos.permute(0, 2, 1, 3, 4)
    denormalized_video = denormalize_video(videos[0])
    sampled_video_every_other_frame = denormalized_video[::2].detach().cpu().permute(0, 2, 3, 1).numpy()
    sampled_video_every_other_frame = rescale_pixel_values(sampled_video_every_other_frame)
    frames_list = [Image.fromarray(frame) for frame in sampled_video_every_other_frame]
    frames_list[0].save(f"{output_dir}vid_{step}.gif", save_all=True, append_images=frames_list[1:], duration=10, loop=0)

def find_optimal_assignment(scores, epsilon, sinkhorn_iterations, world_size=1):
    """
    Computes the Sinkhorn matrix Q.
    :param scores: similarity matrix
    :return: Sinkhorn matrix Q
    """
    with torch.no_grad():
        q = torch.exp(scores / epsilon).t()
        q = sinkhorn(q, sinkhorn_iterations, world_size=world_size)
        # q = torch.softmax(scores / epsilon, dim=0)
        # q = q / q.sum(dim=1, keepdim=True)
    return q

def load_mlp_weight(path, model):
    
    state_dict = torch.load(path, map_location="cpu")
    state_dict = state_dict['model']
    filtered_state_dict = {k.replace('head.', ''): v for k, v in state_dict.items() if 'head.' in k and 'decoder' not in k}

    model.head.load_state_dict(filtered_state_dict, strict=True)
    
    # set require gradients false for this module
    for param in model.head.parameters():
        param.requires_grad = False
        print("Fixed MLP weights")
    return model