StyleGAN2 Distillation

Together with my brother, we re-implemented the work of Evgeny et. al.in entitled 'StyleGAN2 Distillation for Feed-forward Image Manipulation'. In their work, the authors show that a Pix2PixHD network can be trained on a synthetic dataset generated by a GAN (StyleGAN2).

In the re-implementation provided, we focus on the style-mixing task, and manage to train a vanilla Pix2PixHD network into creating a style-mixed face starting from 2 pictures of real persons.

The latest checkpoints of the network can be found here.
The notebook containing the re-implementation of the project is publicly available at this link.
Documentation of how the project was reimplemented can be found at the following link.