Project Goals

• a practice on training a U-GAT-IT model to generate unpaired data
• self-implemented dataset, add Asian faces to the existing one
• Based on Shanshui, doing ink painting for face. ( or outputs that feature an artistic face style of Asian culture, ex claborate-style painting(工笔细描), the Dunhuang frescoes(敦煌湿画). )
• evaluation (1. use the current dataset vs my dataset; 2. apply other GAN to evaluate the performance; 3. qualitative analysis with survey.)

Current Problem: data emphasizing on whiten people

Methods

GAN / U-GAT-IT, unpaired image-to-image training.

Platforms: Google colab(planning). (Tom is back!!!!✌️， forward the it email

Quantitative and qualitative analysis.

Home Community

Graphics, Machine Learning, GAN, Asian culture generative arts

Related Works

How tiktok used GAN on creating face filters

Few-shot Knowledge Transfer for Fine-grained Cartoon Face Generation

minivision-ai/photo2cartoon

U-GAT-IT