train problem

[O-Y-ZONE]

when I was training your code, I found the loss value became the nan. And the PSNR value also became the nan. Moreover, the best model didn't change after the 10 epoch.
Could you provide the completed train code or help me to solve this problem? Thank you sincerely!

[Tombs98]

when I was training your code, I found the loss value became the nan. And the PSNR value also became the nan. Moreover, the best model didn't change after the 10 epoch. Could you provide the completed train code or help me to solve this problem? Thank you sincerely!

Sorry for your question, but the details of our training will be published after the paper is published. We haven't had this problem in our training.

[yayaYsmile]

当我训练你的代码时，我发现损失值变成了 nan。而PSNR值也成了楠。此外，最好的模型在 10 个纪元之后没有改变。你能提供完整的火车代码或帮助我解决这个问题吗？衷心感谢您！
Hello, is the problem solved? I had a similar problem after 200 rounds. Loss suddenly became inf.

[yanghan0617]

你好 训练时 损失一直是负的 梯度是零 请问怎么解决

[Tombs98]

你好 训练时 损失一直是负的 梯度是零 请问怎么解决

#6 (comment)