CPM

Color Makeup - Training Guideline

color-makeup.png
This is a training guideline for Color Branch (C), one out of two branches in Color-Pattern Makeup Transfer (CPM).

For Color Branch, we used the same CycleGAN-based model like BeautyGAN. But instead of normal training pair, we used our novel UV-space.

  1. Download Makeup Transfer Dataset & unzip it. The dataset folder should be like:

     all
 |__ images
     |__ makeup
     |__ non-makeup
 |__ segs
     |__ makeup
     |__ non-makeup

  2. Get UV texture: Use PRNet to get respective uv-map texture of each image and its segmentation mask

python create_beautygan_uv.py --path /path/to/makeup_transfer_dataset --savedir /path/to/save_folder

Result will be saved in --savedir folder. You should find image pair like this:

sample.png
From left to right: Segmentation Mask, UV-texture.
  1. Train Color Branch: Follow instruction at BeautyGAN-pytorch-reimplementation or BeautyGAN-pytorch-implementation to re-train model with newly established dataset.
This site is open source. Improve this page.