Created by Yuval Nirkin.
This library provides video\image sequence functionality for finding face landmarks and bounding boxes.
Main features:
- Matlab interface.
- Saving and loading per sequence.
- Face tracking across frames in a sequence.
- Face statistics for finding the most dominant face.
This library is released as part of the face frontalization project: http://www.openu.ac.il/home/hassner/projects/frontalize
Link for the demonstration video:
There are 3 ways to use the library:
- Matlab interface. Please take a look at the MATLAB Tutorial.
- C++ interface. Please take a look at the doxygen generated documentation.
- Command line tools. Use --help for more information on each tool.
Library | Minimum Version | Notes |
---|---|---|
Boost | 1.47 | |
OpenCV | 3.0 | |
dlib or dlib (Windows) | 18.18 | |
vsal | 1.0 | |
OpenCV's extra modules | 3.0 | Optional - For the LBP face tracker |
protobuf | 3.0.0 | Optional - For loading and saving |
Matlab | 2012a | Optional - For building the MEX function |
- Use CMake and your favorite compiler to build and install the library or download the available binaries from here.
- Add find_face_landmarks/bin to path.
- Add find_face_landmarks/interfaces/matlab to Matlab's path
- Download the landmarks model file: shape_predictor_68_face_landmarks.dat
If you find this code useful, please make sure to add suitable references to the original DLIB library and the frontalization paper. Bib items for both are:
DLIB
@article{king2009dlib,
title={Dlib-ml: A machine learning toolkit},
author={King, Davis E},
journal={J. Mach. Learning Research},
volume={10},
pages={1755--1758},
year={2009},
publisher={JMLR. org}
}
Frontalization
@inproceedings{HHPE:CVPR15:frontalize,
author = {Tal Hassner and Shai Harel and Eran Paz and Roee Enbar},
title = {Effective Face Frontalization in Unconstrained Images},
booktitle = {IEEE Conf. on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2015},
URL = {Available:~http://www.openu.ac.il/home/hassner/projects/frontalize}
}