igul222 / improved_wgan_training

Code for reproducing experiments in "Improved Training of Wasserstein GANs"
MIT License
2.35k stars 668 forks source link

Questions about the loss #85

Open IronLavender opened 5 years ago

IronLavender commented 5 years ago

Hello, I don't know how to analyse the loss for generator and discriminator correctly. When I run on mnist, the two losses are negative. Should we want the absolute value of the discriminator loss to fluctuate around 0.5? And what about the generator loss?