wangsssky / yolo_underwater Goto Github PK

Hi, I've been playing around with your code. Unfortunately the function build_targets which you use in each YOLOLayer sometimes goes out of bounds when building the obj_mask. It all starts from this line, apparently:

Convert to position relative to box
target_boxes = target[:, 2:6] * nG
If I understood your intention correcly, you are indexing each box but the indexing appears to be non-zero based. That's why at this line, at Tensor position 2 it might go out of bounds because gj = nG = obj_mask.shape[2]

obj_mask[b, best_n, gj, gi] = 1