gan training

Question

gan training

Closed this issue 6 years ago · 5 comments

Did you employ WGAN for training? I notice this clip_grad_norm_ in your code.

Answer 1 · 2019-01-13T08:39:26.000Z

We don't employ WGAN in this code. The reason for applying gradient clipping is to avoid explosive gradient and thus stabilize the training.

Answer 2 · 2019-01-13T09:07:26.000Z

@hytseng0509 Yes, the training process seems healthy using gradient clipping. Another question is what could we know with the GAN loss, or say, what should the disContent loss look like if we finish training the model.

Answer 3 · 2019-01-13T21:56:55.000Z

We usually stop the training when the quality of the translated images stop improving or G_loss converges.

Answer 4 · 2019-09-18T16:05:52.000Z

Why is gradient clipping applied only for content discriminator gradients?

Answer 5 · 2019-09-18T16:18:09.000Z

We found it stabilize the training of the content discriminator.