Comments by "Chrysippus" (@4.0.4) on "Generative Adversarial Networks (GANs) - Computerphile" video.

  1. What about using two discriminators for different (but similar) things with the same generator, and in the noise input, using one point in the noise as an indicator of what you want? (The generator's input is all noise except one value) Then later, you could have a slider where you decide how much of either you want. You could even use face recognition to train it on known facial features and landmarks.
    1