Why BCELoss for xy?

[ccqu]

Hi,
I think the original loss for xy is L2, so why in your code is BCELoss used?
https://github.com/DeNA/PyTorch_YOLOv3/blob/master/models/yolo_layer.py#L192
Thanks in advance!

[LukasStruppek]

Hi,
I am having the same question. The paper just mentioned BCE Loss for class prediction.

[YHRen]

add me to the list :D
same question here.