Andy

Also I train from scratch, but the loss is nan for VOC dataset. Here is an example , I wonder whether the loss will be normal after training a long...