the loss is nan

[bbbeach]

I train the yolov3 model with voc2007 and voc 2012.In the second epoch,the loss turn up nan

[Jianhua Yang]

Could you tell what your batch size is？

[bbbeach]

my batch size is 16, using the original learning rate with warmup strategy

[Jianhua Yang]

Ok, I see. I will check the nan problem. Before that, you can try my another YOLO project: https://github.com/yjh0410/yolov2-yolov3_PyTorch. @bbbeach

[bbbeach]

thanks