Correct loss in GAN implementation?

timsainb / tensorflow2-generative-models

Implementations of a number of generative models in Tensorflow 2. GAN, VAE, Seq2Seq, VAEGAN, GAIA, Spectrogram Inversion. Everything is self contained in a jupyter notebook for easy export to colab.

996 stars 187 forks source link

Correct loss in GAN implementation? #8

Closed zjost closed 5 years ago

zjost commented 5 years ago

In this notebook I noticed the following line in the GAN.compute_loss method: gen_loss = gan_loss(logits=logits_x_gen, is_real=True). Shouldn't is_real=False?

zjost commented 5 years ago

Closing, as I'm seeing now this is the intended implementation, which is a heuristic objective meant to provide stronger gradients to the Generator.