GAN on Traditional Chinese Painting

2022 Spring

Motivation

The project is inspired by the work of Alice Xue, paper in WACV 2021.

 

 In recent years, AI art has been one of the newest developments in the field with a variety of cutting-edge tools. Nowadays, AI art can transfer the style of one image to another, create new paintings, restore old pictures, improve the resolution of videos, etc. With that, and to some extent, AI has proven its ability to be creative via generating new samples that have not been seen before.  Therefore, we aim in this project to study several deep learning models to create new Chinese paintings. We plan to improve the traditional AI art architecture based on Generative Adversarial Network (GAN).

I Implemented different GAN variants on Chinese paintings and tried stable diffusion on style transfer.

        • DCGAN
        • WGAN-GP
        • CycleGAN
        • SAPGAN: Sketch-And-Paint GAN

The full project can be found here.

DCGAN

WGAN-GP

CycleGAN - style transfer & coloring

SAPGAN: Sketch-And-Paint GAN

Stable Diffusion - style transfer