search

github-pengge / PyTorch-progressive_growing_of_gans

PyTorch implementation of Progressive Growing of GANs for Improved Quality, Stability, and Variation.
580 stars 97 forks source link

About the loss function of LSGAN #6

Open zhangqianhui opened 6 years ago

zhangqianhui commented 6 years ago

You code about the D_loss

d_adv_loss = self.compute_adv_loss(self.d_real, True, 0.5) + self.compute_adv_loss(self.d_fake, False, 0.5)

The official code of D_loss

 if type == 'lsgan': 
        G_loss = L2(fake_scores_out, 0)
        D_loss = L2(real_scores_out, 0) + L2(fake_scores_out, 1) * L2_fake_weight

The value of L2_fake__weight is 0.1.

So, Do you abserve it?

github-pengge commented 6 years ago

I didn't notice that. Thanks.