I'm not a GAN expert, but I have a problem and I would like to understand if GAN could help me in some way. Essentially my problem is to convert a 3D grayscaled volume in another 3D grayscaled volume, where two volumes contain the same total sum (this conservation is essential).
My inputs for an eventual forward pass in the generator are the volume V1, and an image that comes from the same original data that generates V1. So as far I understand normally in GAN generator does not have additional input when separated by discriminator after training. In the training phase, I have obviously V1 and such image, and the relative labels, V2 for the volume and the label image. My intention is to generate for a brand new V1 the right counterpart V2. Does a GAN represent a good approach to tackle that problem? I came to this idea just because CNNs normally do not use to generate but for classifying. Probably I could instruct a 3D CNN to produce a volume, but the GAN approach that basically model a distribution could be more efficient.