- CPM is a holistic makeup transfer framework that outperforms previous state-of-the-art models on both light and extreme makeup styles.
- CPM consists of an improved color transfer branch (based on BeautyGAN) and a novel pattern transfer branch.
- We also introduce 4 new datasets (both real and synthesis) to train and evaluate CPM.
📢 New: We provide ❝Qualitative Performane Comparisons❞ online! Check it out!
![]() |
---|
CPM can replicate both colors and patterns from a reference makeup style to another image. |
Details of the dataset construction, model architecture, and experimental results can be found in our following paper:
@inproceedings{m_Nguyen-etal-CVPR21,
author = {Thao Nguyen and Anh Tran and Minh Hoai},
title = {Lipstick ain't enough: Beyond Color Matching for In-the-Wild Makeup Transfer},
year = {2021},
booktitle = {Proceedings of the {IEEE} Conference on Computer Vision and Pattern Recognition (CVPR)}
}
Please CITE our paper whenever our datasets or model implementation is used to help produce published results or incorporated into other software.
We introduce ✨ 4 new datasets: CPM-Real, CPM-Synt-1, CPM-Synt-2, and Stickers datasets. Besides, we also use published LADN's Dataset & Makeup Transfer Dataset.
CPM-Real and Stickers are crawled from Google Image Search, while CPM-Synt-1 & 2 are built on Makeup Transfer and Stickers. (Click on dataset name to download)
Name | #imgs | Description | - |
---|---|---|---|
CPM-Real | 3895 | real - makeup styles | ![]() |
CPM-Synt-1 | 5555 | synthesis - makeup images with pattern segmentation mask | ![]() |
CPM-Synt-2 | 1625 | synthesis - triplets: makeup, non-makeup, ground-truth | ![]() |
Stickers | 577 | high-quality images with alpha channel | ![]() |
Dataset Folder Structure can be found here.
By downloading these datasets, USER agrees:
- to use these datasets for research or educational purposes only
- to not distribute the datasets or part of the datasets in any original or modified form.
- and to cite our paper whenever these datasets are employed to help produce published results.
- python=3.7
- torch==1.6.0
- tensorflow-gpu==1.14
- segmentation_models_pytorch
# clone the repo
git clone https://github.com/VinAIResearch/CPM.git
cd CPM
# install dependencies
conda env create -f environment.yml
- Download CPM’s pre-trained models: color.pth and pattern.pth. Put them in
checkpoints
folder.
mkdir checkpoints
cd checkpoints
wget https://public.vinai.io/CPM_checkpoints/color.pth
wget https://public.vinai.io/CPM_checkpoints/pattern.pth
- Download [PRNet pre-trained model] from Drive. Put it in
PRNet/net-data
➡️ You can now try it in Google Colab
# Color+Pattern:
CUDA_VISIBLE_DEVICES=0 python main.py --style ./imgs/style-1.png --input ./imgs/non-makeup.png
# Color Only:
CUDA_VISIBLE_DEVICES=0 python main.py --style ./imgs/style-1.png --input ./imgs/non-makeup.png --color_only
# Pattern Only:
CUDA_VISIBLE_DEVICES=0 python main.py --style ./imgs/style-1.png --input ./imgs/non-makeup.png --pattern_only
Result image will be saved in result.png
As stated in the paper, the Color Branch and Pattern Branch are totally independent. Yet, they shared the same workflow:
-
Data preparation: Generating texture_map of faces.
-
Training
Please redirect to Color Branch or Pattern Branch for further details.
🌿 If you have trouble running the code, please read Trouble Shooting before creating an issue. Thank you 🌿
-
[Solved]
ImportError: libGL.so.1: cannot open shared object file: No such file or directory
:sudo apt update sudo apt install libgl1-mesa-glx
-
[Solved]
RuntimeError: Expected tensor for argument #1 'input' to have the same device as tensor for argument #2 'weight'; but device 1 does not equal 0 (while checking arguments for cudnn_convolution)
Add CUDA VISIBLE DEVICES before .py. Ex:CUDA_VISIBLE_DEVICES=0 python main.py
-
[Solved]
RuntimeError: cuda runtime error (999) : unknown error at /opt/conda/conda-bld/pytorch_1595629403081/work/aten/src/THC/THCGeneral.cpp:47
sudo rmmod nvidia_uvm sudo modprobe nvidia_uvm
docker build -t name .