About the size of the detected bounding box in the dataset you offered

[ArchNew]

I've noticed that the detected bboxes seem to be larger than normal ones, 1.25 times larger in the x and y axis to be precise. After a careful exam of your dataloader codes, I found that your process only resizes the bbox ratio to 3:4, not expanding the bbox 1.25 times in scale. I'm assuming the boxes in the dataset you offered have already been enlarged.

My question is, did you simply enlarge all detected bboxes in the dataset you offered, or you also did other things? I want to run your model in other datasets, so I wish I can get the details on how you process the detected bbox. Thanks!