Multi-pose 2D and 3D Face Alignment and Tracking
The face boxes and five facial landmarks within the annotation files are predicted by our face detector (RetinaFace), which achieves state-of-the-art performance on the WiderFace dataset. We have released this face detector, thus the face alignment algorithms can be tested from scratch under in-the-wild environment.
300W dropbox
COFW dropbox
Menpo2D dropbox
MultiPIE dropbox MultiPIE-3D dropbox
XM2VTS dropbox
FRGC dropbox
68/39 landmarks (The landmark configurations are from MultiPIE.)
(1) 300W/Train (68; 3702)
(2) Menpo2D/Train/image/semifrontal (68; 6653)
(3) Menpo2D/Train/image/profile (39; 2290)
(1) 300W/Validation (68; 135)
(2) COFW (68; 507)
(3) 300W/Test (68; 600)
(4) Menpo2D/Test/image/semifrontal (68; 5335)
(5) Menpo2D/Test/image/profile (39; 1946)
(1) 300VW
(1) 300VW
IBUG just provides the landmark annotations, but some face images are from other works. Please cite the original papers first and follow their data license.
@article{deng2018menpo,
title={The Menpo benchmark for multi-pose 2D and 3D facial landmark localisation and tracking},
author={Deng, Jiankang and Roussos, Anastasios and Chrysos, Grigorios and Ververas, Evangelos and Kotsia, Irene and Shen, Jie and Zafeiriou, Stefanos},
journal={International Journal of Computer Vision},
pages={1--26},
year={2018},
publisher={Springer}
}
@inproceedings{zafeiriou2017menpo2d,
title={The menpo facial landmark localisation challenge: A step towards the solution},
author={Zafeiriou, Stefanos and Trigeorgis, George and Chrysos, Grigorios and Deng, Jiankang and Shen, Jie},
booktitle={Computer Vision and Pattern Recognition (CVPR) Workshops},
year={2017}
}
@inproceedings{zafeiriou2017menpo3d,
title={The 3d menpo facial landmark tracking challenge},
author={Zafeiriou, Stefanos and Chrysos, Grigorios and Roussos, Anastasios and Ververas, Evangelos and Deng, Jiankang and Trigeorgis, George},
booktitle={International Conference on Computer Vision (ICCV) Workshops},
year={2017}
}