Pad the feature maps with 0 :param f: [torch.tensor] or torch.tensor

Return a list of precision/recall based on different confidence thresholds

Remove boxes that are not visible (out of image boundary) after motion augmentation

Reset the status of ROI Heads

Reset the track pool

Resume a dormant track

odd-even order swap of the feature tensor in the batch dimension

Return a new track id, when starting a new track

Suspend an active track, and add it to dormant track pools

Update the latest position (bbox) / search region / template feature for each track in the cache

depthwise cross correlation

A data transformation wrapper for video :param video: a list of images :param target: a list of BoxList (per image)

Check each entity whether it is valid or should be filtered (ignored). :param entities: A list of entities (for a single fram

:param iter: :param loss_dict: :param images: :return:

:param search_expansion: expansion ratio (of the search region) w.r.t the size of tracking targets :param min_search_wh: mini

Arguments: output_size (list[tuple[int]] or list[int]): output size for the pooled region scales (list[float]): scale

Arguments: score_thresh (float) nms (float) detections_per_img (int) box_coder (BoxCoder)

:param dataset: the ingested dataset with GluonCVMotionDataset :param sampling_interval: the temporal stride (in ms) of sliding windo

:param dataset: the ingested dataset with COCO-format :param transforms: image transformation :param frames_per_image: how ma

Construct a data loader for inference :param video: a video stream in DataSample format :param clip_len: the length of video

:param iou_thresh: a predicted entity which overlaps with any ignored gt entity with at least iou_thresh would be filtered

:param video_file: video file path :param frame_idxs: frame that are to be processed, a list of integers