got IndexError
jjxyai opened this issue · comments
dlpacklock commented
boxes = box_head[a, :, y, x]
IndexError: tensors used as indices must be long, byte or bool tensors
https://github.com/NVIDIA/retinanet-examples/blob/main/odtk/box.py#L299
i wonder why use '/' instead of '//' to get the index?
i modified the code to
`
# classes = indices // width // height % num_classes # modified
# classes = classes.type(all_cls_head.type())
# # Infer kept bboxes
# x = indices % width # torch.int64
# y = indices // width % height # modified
# a = indices // width // height // num_classes # modified
`
ankushgupta7 commented
I agree, I faced the same problem and it makes sense to modify the code as you have documented above. Single '/' in python can give float values, but '//' will give the integer value. After the modification, It's working fine.