Image to Image Translation with CycleGAN

I use CycleGAN to develop an image-to-image translation system that converts real human faces in selfies to anime faces and vice versa. I use the sample unpaired dataset selfie2anime. This source contains 3400 images each for selfies and anime faces in the training set and 100 in the validation. The selfie-to-anime generator is able to learn important features and make tangible changes under ideal conditions within 5 epochs, but the anime-to-real generator is much harder to train.

Learn more about this on the Github.