killsking / scgan

SCGAN

Implementation of CVPR 2021 paper "Spatially-invariant Style-codes Controlled Makeup Transfer"

Prepare

The pre-trained model is avaiable at https://drive.google.com/file/d/1t1Hbgqqzc_rV5v3gF7HuJ-xiuEVNb8sh/view?usp=sharing.

vgg_conv.pth:https://drive.google.com/file/d/1JNrSVZrK4TfC7pFG-r7AOmGvBXF2VFOt/view?usp=sharing

Put the G.pth and VGG weights in "./checkpoints" and "./" respectively.

Environments:python=3.8, pytorch=1.6.0, Ubuntu=20.04.1 LTS

Train

Put the train-list of makeup images in "./MT-Dataset/makeup.txt" and the train-list of non-makeup images in "./MT-Dataset/non-makeup.txt"

Use the "./scripts/handle_parsing.py" to convert the origin MT-Dataset's seg labels

Use python sc.py --phase train to train

Test

1.Global Makeup Transfer

python sc.py --phase test

2.Part-specific Makeup Transfer

python sc.py --phase test --partial

3.Global Interpolation

python sc.py --phase test --interpolation

4.Part-specific Interpolation

python sc.py --phase test --partial --interpolation