NaN Loss Issue During LOLv2-syn Dataset Training

shermanlian / spatial-entropy-loss

Equipping Diffusion Models with Differentiable Spatial Entropy for Low-Light Image Enhancement, CVPRW 2024. Best LPIPS in NTIRE chanllenge.

https://arxiv.org/abs/2404.09735

MIT License

15 stars 1 forks source link

NaN Loss Issue During LOLv2-syn Dataset Training #4

Open RenzhiHu111 opened 1 month ago

RenzhiHu111 commented 1 month ago

Hello, when training the LOLv2-syn dataset, the loss keeps becoming NaN, and adding a small constant to sqrt doesn't solve the issue. Switching from AdamW to Lion doesn't reduce the loss either.