2022-04-02 Update: If you could not download the weights with LFS, check out issue #7 (comment) for alternative downloading links.

2022-03-04 Update: We have released the FP-Age model which can perform face parsing and age estimation simultaneously, please visit https://github.com/ibug-group/fpage for details.

Official repo for our paper RoI Tanh-polar transformer network for face parsing in the wild.

Note: If you use this repository in your research, we kindly rquest you to cite the following paper:

@article{lin2021roi,
title = {RoI Tanh-polar transformer network for face parsing in the wild},
journal = {Image and Vision Computing},
volume = {112},
pages = {104190},
year = {2021},
issn = {0262-8856},
doi = {https://doi.org/10.1016/j.imavis.2021.104190},
url = {https://www.sciencedirect.com/science/article/pii/S0262885621000950},
author = {Yiming Lin and Jie Shen and Yujiang Wang and Maja Pantic},
keywords = {Face parsing, In-the-wild dataset, Head pose augmentation, Tanh-polar representation},
}

• git-lfs
• Numpy: $pip3 install numpy
• OpenCV: $pip3 install opencv-python
• PyTorch: $pip3 install torch torchvision
• ibug.roi_tanh_warping: See this repository for details: https://github.com/ibug-group/roi_tanh_warping.
• ibug.face_detection (only needed by the test script): See this repository for details: https://github.com/hhj1897/face_detection.

git clone https://github.com/hhj1897/face_parsing
cd face_parsing
git lfs pull
pip install -e .

python face_warping_test.py -i 0 -e rtnet50 --decoder fcn -n 11 -d cuda:0

Command-line arguments:

-i VIDEO: Index of the webcam to use (start from 0) or
          path of the input video file
-d: Device to be used by PyTorch (default=cuda:0)
-e: Encoder (default=rtnet50)
--decoder: Decoder (default=fcn)
-n: Number of facial classes, can be 11 or 14 for now (default=11)

The training and testing images, bounding boxes, landmarks, and parsing maps can be found in the following:

• Google Drive
• ibug link

Label map for 11 classes:

0 : background
1 : skin (including face and scalp)
2 : left_eyebrow
3 : right_eyebrow
4 : left_eye
5 : right_eye
6 : nose
7 : upper_lip
8 : inner_mouth
9 : lower_lip
10 : hair

Label map for 14 classes:

0 : background
1 : skin (including face and scalp)
2 : left_eyebrow
3 : right_eyebrow
4 : left_eye
5 : right_eye
6 : nose
7 : upper_lip
8 : inner_mouth
9 : lower_lip
10 : hair
11 : left_ear
12 : right_ear
13 : glasses