apply augmentation :param bbox: original bbox in image :param param: augmentation param, shift/scale :param shape: image shape, h, w,

Same as `torch.nn.functional.binary_cross_entropy`, but returns 0 (instead of nan) for empty inputs. `flip_class` inverts the target since th

Concatenates a list of BoxList (having the same image size) into a single BoxList Arguments: bboxes (list[BoxLis

Apply diagonal covaraiance Cauchy-distributed noise to the N,D array x. Parameters: ----------- x : array (N,D) array of v

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

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

Compute the total number of unique clips annotated.

Compute the video hours across all annotations.

Compute the total number of visual queries annotated.

Compute the set of query to response separation in # frames.

Compute the set of response track lengths in # frames.

Compute the set of response track lengths in # frames.

Compute the video hours annotated per scenario. Note: We only count scenarios for which we have more than `thresh` hours of annotations.

Compute the video hours annotated per university.

image - (H, W, 3) numpy array

Center x, y to top-left x, y. Args: bbox: bounding box with [x, y, w, h], x,y are center position of the box. context_ratio:

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

Run SuperGlue on a pair of keypoints and descriptors

Run SuperPoint (optionally) and SuperGlue SuperPoint is skipped if ['keypoints0', 'keypoints1'] exist in input Args: data:

Compute keypoints, scores, descriptors for image

Forward pass with input x.

Forward pass with input x.

:param feat: (H1W1+H2W2, bs, C) :param mask: (bs, H1W1+H2W2) :param query_embed: (N, C) or (N, B, C) :param pos_embe