shaform / deepnetworks

My implementations of deep neural networks for practice.

Hi there 👋

deepnetworks's Issues

Some questions about the loss function of iwacgan and the discriminator settings

Hi your project is very detailed and easy to read. But I am a pytorch user, and can't understand it particularly clear. Therefore，there some question I try to figure out.

If i have not made a mistake, you use the three discriminators in iwacgan to deal with real, fake, the interpolation samples respectively. I have read some iwgan code where only one discriminator is used. What is the difference between them? Will the three discriminators are updated at the same time according the loss function?@shaform

In addition, the generator loss and discriminator loss is
'''python
self.g_total_loss = self.g_loss + self.g_c_loss + self.g_reg_loss

self.d_total_loss = (self.d_loss + self.d_c_loss + self.d_grad_loss+ self.d_reg_loss)
'''
The total loss function looks equal to a value that iwgan loss plus the category crossover entropy loss. Do not need any modification operation?