04 Oct 2019
Lately, I’ve been experimenting with an image-to-image translation GAN called UGATIT. It features some novel techniques that give really cool results. I will likely try to explain how it works in a future post, but for now I just want to share the freaky output.
After one epoch of training (i.e. not a lot) on a dataset of cartoon faces and real faces, the following was produced. The model has experienced a slight mode collapse, but there are still subtle differences between the samples. The faces remind me of Aphex Twin and are frankly horrifying to look at.
The image is quite large so I leave it as a link. Enter at your own risk! Montage of cartoon2real faces.