Fix threshold in sam_predict

[nidefawl]

SAMDetectorSegmented can return incorrect masks in the batch_masks output.

This is because the threshold for the 3 masks returned by SamPredictor predict can differ.

ComfyUI-Impact-Pack/modules/impact/core.py

Lines 463 to 489 in ddaedfe

	def sam_predict(predictor, points, plabs, bbox, threshold):
	point_coords = None if not points else np.array(points)
	point_labels = None if not plabs else np.array(plabs)
	box = np.array([bbox]) if bbox is not None else None
	cur_masks, scores, _ = predictor.predict(point_coords=point_coords, point_labels=point_labels, box=box)
	total_masks = []
	selected = False
	max_score = 0
	for idx in range(len(scores)):
	if scores[idx] > max_score:
	max_score = scores[idx]
	max_mask = cur_masks[idx]
	if scores[idx] >= threshold:
	selected = True
	total_masks.append(cur_masks[idx])
	else:
	pass
	if not selected:
	total_masks.append(max_mask)
	return total_masks

But make_sam_mask_segmented expects 3 masks per detection.

ComfyUI-Impact-Pack/modules/impact/core.py

Line 839 in ddaedfe

return (mask, merge_and_stack_masks(stacked_masks, group_size=3))

This commit fixes it by respecting the threshold value and returning all or none of the predicted masks.

[ltdrdata]

You need to check tensor shape of image.

It is b, h, w, c.