With help from @sgugger and @jcreinhold I managed to get a CycleGAN loading my custom data and training, using only the fastai library. (That’s kind of an objective of mine, since I’d rather focus on a single platform/library, as much as possible.) However, I’m really not sure how to go about generating an output. My application is a “sketch-to-image” type of model, so I want to provide an image from my test set (a single 64x64 image, not a pair), and just get a single “fake” image from the GAN’s generator. It seems like I need something like
cycle_gan.G_B(image.data), but I haven’t been able to get anywhere with that approach.
Any tips greatly appreciated.