tfm.vision.box_ops.bbox_generalized_overlap

Calculates the GIOU between proposal and ground truth boxes.

tfm.vision.box_ops.bbox_generalized_overlap(
    boxes, gt_boxes
)

The generalized intersection of union is an adjustment of the traditional IOU metric which provides continuous updates even for predictions with no overlap. This metric is defined in https://giou.stanford.edu/GIoU.pdf Note, some gt_boxes may have been padded. The returned giou tensor for these boxes will be -1.

boxes a Tensor with a shape of [batch_size, N, 4]. N is the number of proposals before groundtruth assignment (e.g., rpn_post_nms_topn). The last dimension is the pixel coordinates in [ymin, xmin, ymax, xmax] form.
gt_boxes a Tensor with a shape of [batch_size, max_num_instances, 4]. This tensor may have paddings with a negative value and will also be in the [ymin, xmin, ymax, xmax] format.

giou a Tensor with as a shape of [batch_size, N, max_num_instances].