Calculates the IoUs and IoAs between the detection masks and the ground truth masks.
tfm.vision.mask_ops.instance_masks_overlap(
boxes: tf.Tensor,
masks: tf.Tensor,
gt_boxes: tf.Tensor,
gt_masks: tf.Tensor,
output_size: List[int],
mask_binarize_threshold: float = 0.5
) -> Tuple[tf.Tensor, tf.Tensor]
IoU: intersection over union.
IoA: intersection over the area of the detection masks.
Args |
boxes
|
a tensor with a shape of [batch_size, N, 4]. The last dimension is
the pixel coordinates in [ymin, xmin, ymax, xmax] form.
|
masks
|
a float tensor with a shape of [batch_size, N, mask_height,
mask_width] representing the instance masks w.r.t. the boxes .
|
gt_boxes
|
a tensor with a shape of [batch_size, M, 4]. The last dimension is
the pixel coordinates in [ymin, xmin, ymax, xmax] form.
|
gt_masks
|
a float tensor with a shape of [batch_size, M, gt_mask_height,
gt_mask_width] representing the instance masks w.r.t. the gt_boxes .
|
output_size
|
two integers that represent the height and width of the output
masks.
|
mask_binarize_threshold
|
a float representing the threshold for binarizing
mask values. Default value is 0.5.
|
Returns |
iou
|
a tensor with as a shape of [batch_size, N, M].
|