Pytorch-yolov3: In ListDataset, collate_fn in utils/datasets.py , you're resizing the image. Aren't you also supposed to resize the bounding boxes ?
andohuman
All 3 comments
I also have the same doubts about this issue.
SWHL
on 18 Jul 2019
I am also confused about the shape of targets in function collate_fn, it seems to be (batchsize, w, h) where it is used in the following part, but if I just read collate_fn , I think it would be a two dimensional tensor, can someone tell me if I'm wrong? Or should I add unsqueeze(0) at the end of each channel of targets?
EddieEduardo
on 24 Sep 2019
No need to resize because they are normalized between 0-1. Resizing image does not change xy or wh of the bboxes relative to the image.
did85
on 17 Nov 2019
👍3
Was this page helpful?
0 / 5 - 0 ratings
Related issues
nationalflag
路
5Comments
qianwangn
路
3Comments
zhaowt61
路
4Comments
hezhangubc
路
4Comments
mayaghaei
路
4Comments
Most helpful comment
No need to resize because they are normalized between 0-1. Resizing image does not change xy or wh of the bboxes relative to the image.