hi,
As I read the DCGAN tutorial, I have some questions.
When training the generator the gan loss is calculated by log(D(G(x))).
In the backward stage, the model calculates the discriminator gradients and then calculates the generator gradients.
Q1. So I should set all weights parameters to requires_grad = True
in the discriminator, is it?
Q2. If I set the parameters to requires_grad = False
in the discriminator, can the model calculates the gradients in the generator?
Q3. If I add another well-trained classifier in the gan model to calculate loss like C(G(x)), which parameters of requires_grad
in the classifier should I set can that the model calculates the generator gradients?
Thank you!!!