Compute iou matrix between two lists of Entity bbox in AnnoEntity is in the format of xywh Different from boxlist_iou in terms of not ad

Create dummy boxlist, with bbox [-1, -1, -1, -1], id -1, label -1 when num_boxes = 0, it means return an empty BoxList

Duplicate samples in box list by concatenating multiple times.

Get dataset specific filter function list, if there is any

Extend the bounding box to define the search region :param in_box: a set of bounding boxes in previous frame :param min_wh: t

Construct a boxlist with bbox as bboxes, all other fields to be default id -1, label -1

Automatically generate a model name that carries key information about configuration; Those information includes the backbone, functionality

Do the greedy matching across predictions and ground truth annotations Returns the matching ious for every predictions and ground tru

Load GluonCVMotionDataset format annotations for downstream training / testing

Returns BoxList from `boxes` and adds probability scores information as an extra field `boxes` has shape (#detections, 4 * #c

Sample examples

Update the coordinates of bounding boxes in the padded image

Check and load ground-truth labels

post_process the tracks to filter out short and non-confident tracks :param tracks: un-filtered tracks :return: filtered trac

Use box head to refine the bounding box location The final vis score is an average between appearance and matching score

Assign for each proposal a matched id, if it is matched to a gt box Otherwise, it is assigned -1

Adjust the coordinate of the bounding box within image crop specified by (x1, y1, x2, y2)

Convert a list of boxlist to entities :return:

Constructs a new box head. By default, uses ROIBoxHead, but if it turns out not to be enough, just register a new class and make it a par

Computer Average Precision (AP) given a list of score :param scores: A list of confidence scores of detections :param pred_ious: A list

Do inference on a specific video (sample) :param cfg: configuration file of the model :param model: a pytorch model :param sample: a

Wrap up the entity to maskrcnn-benchmark compatible format - BoxList

:param samples: a list of (sample_id, sample:DataSample) :param predicted_samples: a dict with (sample_id: predicted_tracks:DataSample) :

Evaluate the detection performance (COCO-style ap) on PoseTrack dataset :param gt: ground truth annotations for all videos :type gt: dict

Expire the suspended tracks after they are inactive for a consecutive self._max_dormant_frames frames

Get the cache (state) that is necessary for tracking output: (features for tracking targets, search region,

Returns bounding-box detection results by thresholding on scores and applying non-maximum suppression (NMS).

Perform correlation on feature maps and regress the location of the object in other frame :param features: a list of feature maps fro

Generate hard negative pair by sampling non-negative proposals

Generate negative training examples

Generate positive and hard negative training examples

:param vid_class_gt: the ground truths for a specific class, in DataSample format :param vid_class_pred: the predictions for a specific class

Get the corresponding target box given by the 1-off indices if there is no matching target box, it returns a dummy box

Process the long videos to a small video chunk (with self.clip_len seconds) Video clips are generated in a temporal sliding window fa

Get images inpainted with bounding boxes :param images: A batch of images are packed in a torch tensor BxCxHxW :param boxes: