Hi all,
I wrote this GAN (in the example it uses fake input):
Basically, the generative accuracy is quantified in terms of image difference.
The generator net and the discriminator net get trained (individually), but it looks like they don’t “communicate”. I mean, the discriminator judgment doesn’t affect the generative training…the training goes parallel, without affecting each other.
Any idea? Am I missing something?
I multiply the discriminator score by a float factor in order to make it comparable to the image difference.
Thank you in advance.
Best,
Paolo