Conditional GANs

I was asked after class how to create a GAN where you can ask the model to create an image of some particular category. I think this is an interesting enough question that’s it’s worth its own thread!

The answer is to use something called a “conditional GAN”. The probabilistic concept of “conditionals” comes up all the time in deep learning, so it’s well worth familiarizing yourself with it in any case.

Here’s a starting point for anyone who would like to investigate this further: . There hasn’t been any conditional WGANs yet AFAIK, so that would be a new contribution, if anyone were able to get that working…


I was unable to find any useful information about what conditionals really are. In the link you posted, it seems that its just another vector that ‘tell’ the network what to produce. Is that all or is there something more to it ?