Extract multi-scale features using backbone network

Get frames and corresponding masks for a specific action Args: video_id: ID of the video frame_indices: List

3x3 convolution with padding

Compute region similarity as the Jaccard Index. Arguments: annotation (ndarray): binary annotation map. segmentation (ndarray

Get valid frames for specified action in video Args: video_id: video ID action_name: action name category:

Return an activation function given a string

Compute mean,recall and decay from per-frame evaluation. Calculates precision/recall for boundaries between foreground_mask and gt_mask using morp

A wrapper around :function:`torch.nn.functional.grid_sample` to support 3D point_coords tensors. Unlike :function:`torch.nn.functional.grid_s

Initialize `module.weight` using the "XavierFill" implemented in Caffe2. Also initializes `module.bias` to 0. Args: module (torc

Cleanup DDP environment

Sinusoidal position encoding table supporting single sequence

Constructs a ResNet-50 model. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet

Sample frames ensuring at least specified number of action frames, uniformly sampling from entire action sequence Args: valid_fram

From a segmentation, compute a binary boundary map with 1 pixel wide boundaries. The boundary pixels are offset by 1/2 pixel towards the origin f

Initialize evaluation metrics

Args: x: (B, D, H, W, C) window_size (tuple[int]): window size Returns: windows: (B*num_windows, window_size*window_

Args: x: (B, H, W, C) window_size (int): window size Returns: windows: (num_windows*B, window_size, window_size, C)

Add index to tasks.

Build backbone network

We estimate uncerainty as L1 distance between 0.0 and the logit prediction in 'logits' for the foreground class in `classes`. Args:

Efficient version of torch.cat that avoids a copy if there is only a single element in a list

Cleanup DDP environment

Compute the DICE loss, similar to generalized IOU for masks Args: inputs: A float tensor of arbitrary shape. The pred

Evaluate MoVe inference results.

The parameter require_cls and oracle_mask only used for visualization and ablation study Please dont use them during normally trainin

Parse command line arguments

Get all evaluation metrics

Get similar actions list for given action Args: action: target action num_actions: number of similar actions

Sample points in [0, 1] x [0, 1] coordinate space based on their uncertainty. The unceratinties are calculated for each point using 'unce

Create learning rate scheduler with warmup and cosine annealing

Initialize the weights in backbone. Args: pretrained (str, optional): Path to pre-trained weights. Defaults to Non

param: pred of size [N x H x W] param: target of size [N x H x W]

Sample frames ensuring at least specified number of action frames Args: valid_frames: frames containing target action

设置随机种子以确保实验的可重复性，特别是在多GPU训练环境中。 Args: seed (int): 随机种子值，默认为1234 deterministic (bool): 是否使用确定性算法，默认为True

Initialize DDP environment

Initialize DDP environment

Args: inputs: A float tensor of arbitrary shape. The predictions for each example. targets: A float tensor with t

Track the progress of parallel task execution with a progress bar. The built-in :mod:`multiprocessing` module is used for process pools and tasks

Main training function

Convert the model into training mode while keep layers freezed.

Args: windows: (B*num_windows, window_size, window_size, C) window_size (tuple[int]): Window size H (int): Height of imag

Args: windows: (num_windows*B, window_size, window_size, C) window_size (int): Window size H (int): Height of image