Toonify

Repository accompanying the paper:

Resolution Dependent GAN Interpolation for Controllable Image Synthesis Between Domains

Accepted to the Neurips 2020 Machine Learning for Creativity and Design NeurIPS 2020 Workshop.

Datasets

Datasets used in the paper:

Aligned Ukiyo-e Faces

Over five thousand aligned ukiyo-e faces at 1024x1024 pixel resolution. The dataset is released under a CC BY-SA 4.0 license. For more details see the accompanying dataset release blog post.

Download here - V2 (Google Drive)

Cartoon Faces

Download here (Mega)

Code

I'm working on releasing an end to end set of examples but in the mean time please see the following two Colab notebooks:

• Network blending - Demonstrates the layer swapping technique for resolution dependent GAN interpolation
• Toonify yourself - Recreates the right hand panel of Figure 3 from the paper

More

For more details also see these blog posts:

• https://www.justinpinkney.com/stylegan-network-blending/
• https://www.justinpinkney.com/toonify-yourself/
• https://www.justinpinkney.com/making-toonify/

While you're here why not Toonify Yourself you can also try out Toonify HD