Returns the instance specified by the name. Args: name (str): name of the specified instance.

Returns: (float): the total number of seconds since the start/reset of the timer, excluding the time when the

r"""Fills the input Tensor with values drawn from a truncated normal distribution. The values are effectively drawn from the normal distributi

Stop to record time.

Start to record time.

Update the current stats. Args: top1_err (float): top1 error rate. top5_err (float): top5 error rate.

Args: cfg (Config): global config object.

Add a new scalar value to the deque.

Wrapper for downloading the file object. Args: obj_file (str): the target file to be downloaded (if it starts

Calculate the current median value of the deque.

log the stats of the current iteration. Args: cur_epoch (int): the number of current epoch. cur_iter (int): t

Reset the Meter.

Update stats using custom keys. Args: stats (dict): additional stats to be updated.

Generate bbox and apply lambda correction.

Load the checkpoint from the given file. Args: path_to_checkpoint (string): path to the checkpoint to load. model (model): m

Reset the timer.

Get full path with epoch and subset Args: prefix (str): path prefix. epoch (int): epoch number of these proposals tes

With 50% prob, negate the value

Constructs the data loader for the given dataset. Args: cfg (Configs): global config object. details in utils/config.py split

Get path for storing checkpoints. Args: path_to_job (string): the path to the folder of the current job.

Args: cfg (Config): global config object.

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

Performs ResNet style weight initialization. Args: fc_init_std (float): the expected standard deviation for fc layer. zero_in

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

Replace the attributes in the base config by the values in the coming config, unless preserve base is set to True. Args:

Builds the video model. Args: cfg (Config): global config object that provides specifics to construct the model. gpu_id (Opti

Retrieve the loss given the loss name. Args (int): loss_name: the name of the loss to use.

Pause the timer.

Transfer epic annotation strings to time. Args: instr (string): Time string read from annotation files.

Get all predicitons of the given label. Return empty DataFrame if there is no predcitions with the given label.

Join a list of paths. Args: path_list (list): list of paths.

We divide each videos to multi clips, and each clip contains 5s. For example, if the seg_time is [8.5, 16.1], which means that we nee

generate sample mask for each point in Boundary-Matching Map Args: prop_boundary_ratio (float): Boundary expand ratio.

calculate the overlap proportion between the anchor and all bbox for supervise signal, Args: anchors_min (np.ndarry): 1d anchors star

Convert the given RGB Image Tensor to Grayscale. For RGB to Grayscale conversion, ITU-R 601-2 luma transform is performed which is L = R * 0

Computes average precision for each class in the subset.

Args: cfg (Config): the global config object.

Load feature from file. Args: feature_path (str): File path for a feature. num_retries (int): Retry time if d

Preprocess the checkpoints for video vision transformers. Enabled in the cfg. It includes processing the positional embedding by repeating an

From https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/drop.py. Drop paths (Stochastic Depth) per sample (when

Evaluate the model on the val set. Args: val_loader (loader): data loader to provide validation data. model (model): model/mo

Calculate and log the final ensembled metrics. ks (tuple): list of top-k values for topk_accuracies. For example, ks = (1

Since different sliding windows have overlap, we desigin this function to fuse sliding windows. Args: maps_dict (dict): {

Returns crop parameters. Args: speed_factor (float): frac{distance_to_go}{total_distance} total_length (i

Calculate the global mean value.

Get the last checkpoint from the checkpointing folder. Args: path_to_job (string): the path to the folder of the current job.

Given the configs, retrieve the specified lr policy function. Args: lr_policy (string): the learning rate policy to use for the job.

Compute statistics for the current model given the config. Args: model (model): model to perform analysis. cfg (Config): the

Determines if the given directory contains a checkpoint. Args: path_to_job (string): the path to the folder of the current job.

Compute jaccard score between a box and the anchors. Args: anchors_min (np.ndarry): 1d anchors start position, shape is N. an

Log the stats of the current epoch. Args: cur_epoch (int): the number of current epoch.

Perform multi-view test on the specified test set, where {cfg.TEST.NUM_ENSEMBLE_VIEWS} clips and {cfg.TEST.NUM_SPATIAL_CROPS} crops are sampl

Perform sliding windows test on the specified test set Args: test_loader (loader): video testing loader. model (model): the p

Given the predictions, labels, and a list of top-k values, compute the number of correct predictions for each top-k value. Args:

r"""Sets the module in training mode. This has any effect only on certain modules. See documentations of particular modules for detai

Automatically decodes parameters for 3D convolution blocks according to the config and its index in the model. Args: cfg (Config):

Update predictions and labels. Args: preds (tensor): model output predictions. labels (tensor): labels.

Args: cfg (Config): global config object.

Args: cfg (Config): global config object.

Args: cfg (Config): global config object.

Apply boundary mathcing operation for input feature Args: x (tensor): 1D feature for boundary mathcing operation.

Configs the transform for the dataset. For train, we apply random cropping, random horizontal flip, random color jitter (optionally