ENSTA-U2IS-AI / LDU

Latent Discriminant deterministic Uncertainty [ECCV2022]

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Nan loss during training

Waynechen92 opened this issue · comments

Has anyone ever met Nan lossing during training for bts_main?
Any possible solution for that?

Hi, thank you for leaving the comments.
Actually, it appears in the original bts training as well, you can check here to find some suggestions given by the authors of bts.

Thanks for your reply xuanlong, I'm also wondering whether it is correct to have negative loss in training for bts_main

Several common loss functions are indeed non-negative such as MSELoss or L1Loss. But it is fine to have a negative loss value during training, not only in LDU case. You can check here for example.