LMescheder / GAN_stability

Code for paper "Which Training Methods for GANs do actually Converge? (ICML 2018)"
MIT License
917 stars 114 forks source link

Gradient Penalty with Auxiliary classifier #10

Open dougsouza opened 5 years ago

dougsouza commented 5 years ago

Gradient penalty claculation should change when D has an auxiliary classifier or the gradient penalty should only be calculated over the real/fake output? Is there any intuition behind of whatever is the best solution?

Thanks!

Johnson-yue commented 5 years ago

@dougsouza do you implement it??

dougsouza commented 5 years ago

@Johnson-yue, yes. I tried gradient penalty only in the real/fake output of the network. But I didn't pursue much, I went back to spectral normalization.

Johnson-yue commented 5 years ago

Maybe spectral normalization is in common use