When I run the SN-GAN notebook it seems to get stuck and never really improves:
Step 500: Generator loss: 0.6939174304008486, discriminator loss: 0.6967933417558673.
ā¦
Step 16500: Generator loss: 0.6933934584856034, discriminator loss: 0.6932449789047234
The images the GAN produces look not far from random noise.
What parameters should I be changing to make the generator model improve?
Hi @ya_coursera_student,
The SN-GAN example is fairly slow to start producing reasonable results, but if you run the notebook in Coursera without any changes, and run it all the way to the end of the 50 epochs, you should start to see some digits forming. I just gave it a try, and this is what I saw after 50 epochs:
