I have an idea (something like, given a certain angle of something, produce an image of that thing from a different angle).
Using a DCGAN seems to be the clear choice, but I don’t understand how to tell the GAN to perform something on some real image.
It seems that the example Jeremy provided uses random noise.
You could try a conditional GAN:
Thanks for that!
My main question is , does a generator only take noise ? or can it learn mappings between arbitrary things ?
For e.g. What about DiscoGAN ? it takes a shoe outline, and produce a coloured image of the shoe. Is the input to that GAN the outline ?