I was looking at the GAN tutorial here and also the example given for NetGANOpereator, and I noticed that in both examples, an unconditional GAN is being trained. This means that there is no control over the class of the images being generated. Instead, I want to train a conditional GAN, explained here and here, whose generator takes as input both a noise vector and also the class to generate (e.g., if given the digit 2 when trained on MNIST, it generates images of 2's). I was wondering whether it is possible to do this in the Wolfram language and how?
Thanks.