Constructs the dataset according to the global config object. Currently supports reading from csv, json and txt. Args:

Transfer string to time or label.

We divide each videos to multi-clips, and each clip contains 5 seconds. This function is used to generate video clips path for easy

Init annotation files and video-duration list for evaluation.

from https://github.com/openai/CLIP/blob/main/clip/clip.py

Get the maps for boundary matching network. Args: gt_bbox (np.array): The shape is N*2. Normalized temporal bounding boxe

Get full bmn training labels. Args: gt_time (np.array): The shape is N*2. The ground truth temporal bounding boxes.

Returns the list for the dataset. Returns: name (str): name of the list to be read

We divide each videos to multi clips, and each clip contains 5s. This function is used to get clips which contain seg_time.

Return a process group based on gloo backend, containing all the ranks The result is cached. Returns: (group): pytorch dist group

generate sample mask for a boundary-matching pair Args: seg_xmin (float): Start time of the proposal. seg_xma

Return a dummy input for model analysis with batch size 1. The input is used for analyzing the model (counting flops and activations etc.

Get annotation and data path for training and validation. Args: index (int): the index of the data in the self._sa

Returns the sample info corresponding to the index. Args: index (int): target index Returns: sa

Get the start and end sequences. Args: gt_bbox (np.array): The shape is N*2. Normalized temporal bounding boxes.

Get a randomized transform to be applied on image. Arguments are same as that of __init__. Arg: T (int): number of fram

Returns the list of frame indexes in the video for decoding. Args: vid_length (int): video length clip_i

Reads ground truth file, checks if it is well formatted, and returns the ground truth instances and the activity classes. Paramete

Reads prediction file, checks if it is well formatted, and returns the prediction instances. Parameters ----------

Init temporal tools for boundary matching map generation. Such as anchors and valid iou mask for boundary matching network.

When loading config for the first time, base config is required to be read.

Initializes the oss bucket. Currently supporting two OSS accounts.

Load feature data and classification data. Args: sample_info (dict): feat_path for feature and cls_path for classificatio

Load the true length of videos for sliding windows. Args: cfg (Configs): global config object. details in utils/config.py

Load the specified yaml file. Args: args: parsed args by `self._parse_args`. file_name (str): the file name t

Merge cfg from command. Currently only support depth of four. E.g. VIDEO.BACKBONE.BRANCH.XXXX. is an attribute with depth of four.

Norm ground truth time into [0,1]. Args: gt_time (np.array): The shape is N*2. There have N ground truth action in seg_ti

Padding all the tensors from different GPUs to the largest ones. Args: tensor (tensor): tensor to pad. group (group): pytorch

Wrapper for argument parser.

Set transformation parameters if required.

Generate sliding windows and matching their ground truth for training and validation. Args: cfg (Configs): global config

Wrapper for downloading the image and generating the PIL.Image object for reading the image. Args: path (str): im

Wrapper for downloading the video and generating the VideoReader object for reading the video. Args: video_path (str

Registers the instance. Args: name (str): name of the entry entry (): instance of the entry, could be any t

Generates the frame index list using segment based sampling. Args: vid_length (int): the length of the whole video

Seriialize the tensor to ByteTensor. Note that only `gloo` and `nccl` backend is supported. Args: data (data): data to be ser

Suppresses printing from the current process.

Set the dict to be attributes of the config recurrently. Args: cfg_dict (dict): the dictionary to be set as the attribute

Adjust hue of an image. The image hue is adjusted by converting the image to HSV and cyclically shifting the intensities in the hue channel

Builds a dataset according to the "dataset_name". Args: dataset_name (str): the name of the dataset to be constructed. cf

Entry point to registered self-supervised learning methods. Returns transformed frames and the self-supervised label. Args:

Initialize `module.weight` using the "MSRAFill" implemented in Caffe2. Also initializes `module.bias` to 0. Args: module (torch.n

Compute average precision (detection task) between ground truth and predictions data frames. If multiple predictions occurs for the same predi

Configures the transformation applied to the pseudo motion sequence.

Evaluates a prediction file. For the detection task we measure the interpolated mean average precision to measure the performance of a

Evaluate action detection performance. Args: video_anno (str): Annotation file path. detection_result_file (str): The detecti

Standardize the frame size according to the settings in the cfg. Args: frame (Tensor): a single frame with the shape o

Simply fuse verb map and noun map for action classification Args: cfg (Config): the global config object. verb_map (dict): ve

Generates labels for joint prediction.

Generates labels for separate prediction.

Returns the read arguments.