FusionYu

GitHub didn't prompt the message. Sorry I didn't reply. I don't have NSST-DualPCNN code. Maybe I won't update this project either. Thank you

I'm not quite sure what the issue you mentioned is caused by. It could be an issue with the training set, or it could be that it hasn't converged yet....

If that still doesn't work, I recommend disabling some of the loss functions and testing the training performance with the basic loss function first.

之前忘记回复了，这份代码已经相当古老了。我建议用tensorboard 来打印损失。或者简单点，直接.item()来打印。 TNO数据，报错是什么？我不太记得这个代码和图像怎么处理的了。如果是尺寸不对，我建议resize或者裁减到某些倍数尺寸。比如640x640。

关闭Resnet18的torch vision 权重加载，极大的概率减小训崩的情况。