Open liangxuy opened 4 years ago
Hi @liangxuy and @jz462
I also encountered this problem during training on pytorch-1.0 branch. I am wondering whether it matters or not, as most of the training losses (as showed in screenshot above, loss_cls_ttl, loss_contrastive_sbj, loss_contrastive_obj, loss_so_contrastive_sbj, etc.) become nan.
Thanks for the code. However, whenever I run the training on VRD dataset with the default setting, I met the 'Warning: NaN or Inf found in input tensor' problem. I wonder if anyone else met similar problem?![image](https://user-images.githubusercontent.com/39529512/65813806-6bf1bb80-e20c-11e9-9db3-bca8b78e0200.png)