rwightman/efficientdet-pytorch

[BUG] Height and width is wrong way around in clip_boxes (transforms.py)

DavidPetrus opened this issue · 2 comments

I think there is a bug when clipping the bounding box annotations. Width and height are switched so Line 57 in transforms.py should be:
clip_upper = np.array([width,height] * 2, dtype=boxes.dtype)

@DavidPetrus The dataset annotations contains bbox in yxyx format by default, do you still believe that to be in error?

Aah I see now, in that case there's no error.