search

YuanXue1993 / SegAN

SegAN: Semantic Segmentation with Adversarial Learning
MIT License
182 stars 58 forks source link

Why is the Prediction results all Nan? #8

Closed hydxqing closed 5 years ago

hydxqing commented 5 years ago

After training, I saved the model parameters of NetS, and then the segmentation prediction is carried out with NetS model. Why is the result all Nan?

YuanXue1993 commented 5 years ago

Maybe try with different learning rate? Are you using your own data?

hydxqing commented 5 years ago

Yes, I use my own data. Loss during training has generally been very small and shows large and small.What's the matter?

YuanXue1993 commented 5 years ago

What do you mean by "shows large and small"? I suggest you start w/o the adversarial loss (e.g., use the generator with cross-entropy loss) to see if there are any bugs in the code. After running smoothly, you can move to the adversarial training.