box_iou

torchvision.ops.box_iou(boxes1: Tensor, boxes2: Tensor, fmt: str = 'xyxy') → Tensor

Return intersection-over-union (Jaccard index) between two sets of boxes from a given format.

Parameters:

• boxes1 (Tensor[..., N, K]) – first set of boxes

• boxes2 (Tensor[..., M, K]) – second set of boxes

• fmt (str) –

  Format of the input boxes. Default is “xyxy” to preserve backward compatibility.

  Supported axis-aligned formats (K=4):

  • 'xyxy': boxes are represented via (x1, y1, x2, y2) corner coordinates.

  • 'xywh': boxes are represented via (x1, y1, width, height).

  • 'cxcywh': boxes are represented via (cx, cy, width, height) center coordinates.

  Supported rotated formats:

  • 'cxcywhr' (K=5): boxes are represented via center, width, height, and rotation angle. (cx, cy) is the center, (w, h) is width and height, r is rotation angle in degrees (counter-clockwise positive).

  • 'xywhr' (K=5): boxes are represented via corner, width, height, and rotation angle. (x1, y1) is the top-left corner, (w, h) is width and height, r is rotation angle in degrees.

  • 'xyxyxyxy' (K=8): boxes are represented via 4 corner coordinates. (x1, y1) is top-left, (x2, y2) is top-right, (x3, y3) is bottom-right, (x4, y4) is bottom-left.

Returns:

the NxM matrix containing the pairwise IoU values for every element in boxes1 and boxes2

Return type:

Tensor[…, N, M]