I'm studying deeply GANs since I'm quite fascinated by them. I'm a pro C++ developer (not in the AI field), so using current AI libs is not a problem (using Python as well). What I was trying to understand is whether I can try to face a pair or Raspi-like boards (with USB TPUs or Coral Dev boards directly) trying to emulate a GAN (I mean for example using one board as discriminator and one as generator, eventually). I'm almost pretty sure my scenario is not so simple to be achieved, considering those machines have not enough powers to handle real large models. However, I would be still interested in a kind of toy model handling, with eventually small data set or so. Eventually, it could also be interesting to stack several of these boards together to improve their training capabilites, but it seems Coral is not suited to do that, working mainly with TensorFlow Lite models only.
Any help would be highly appreciated, tx in advance.