BeautyREC

This repository contains the official implementation of the following paper:

**BeautyREC:Robust, Efficient, and Component-Specific Makeup Transfer. ** CVPRW 2023

Paper Link: paper

Prepare

The pre-trained model is avaiable at "./checkpoints/BeautyREC.pt"

vgg_conv.pth: https://drive.google.com/file/d/1CJ2tk-rfG3ox1hRg_RcohcoVsfaJo41j/view?usp=sharing

Put the VGG weights in "./network/REC/"

Data :

MT dataset: https://drive.google.com/file/d/1jP7CpiczZ9KjTQu87PEERrN7BOrxB5St/view?usp=sharing

Wild dataset:https://drive.google.com/file/d/1bQMglioFb50HVwfYaYhVkFXVJIRf42Nm/view?usp=sharing

Beautyface: https://drive.google.com/file/d/1mhoopmi7OlsClOuKocjldGbTYnyDzNMc/view?usp=sharing

Beautyface parsing(vis): https://drive.google.com/file/d/1sRE-VvC63Cyn_VNUOKYO4WY762_qEjX2/view?usp=sharing

Beautyface parsing maps: https://drive.google.com/file/d/1WgadvcV1pUtEMCYxjwWBledEQfDbadn7/view?usp=sharing

Environments:

python >= 3.6
torch >= 1.0
tensorboardX >= 1.6
utils-misc >= 0.0.5
mscv >= 0.0.3

Train

Put the train-list of makeup images in "./mtdataset/makeup.txt" and the train-list of non-makeup images in "./mtdataset/nomakeup.txt" ( you can randomly split the images.)

train the mt dataset using "./makeuploader/dataset.py"

train the wild dataset using "./makeuploader/wilddataset.py"

You can init the dataset in "./makeuploader/dataloaders.py"

CUDA_VISIBLE_DEVICES=0 python train.py --tag mt --model REC 

Test

Put the test-list of makeup images in "./mtdataset/makeup_test.txt" and the test-list of non-makeup images in "./mtdataset/nomakeup_test.txt"

CUDA_VISIBLE_DEVICES=0 python test.py --model REC  --load checkpoints/BeautyREC.pt

Results