This is the source code for paper

Video-based Remote Physiological Measurement via Cross-verified Feature Disentangling. (Oral)
Xuesong Niu, Zitong Yu, Hu Han, Xiaobai Li, Shiguang Shan, Guoying Zhao
European Conference on Computer Vision (ECCV), 2020.


This code is based on Matlab2018b, Python2.7 and Pytorch 0.4.1


For the VIPL-HR database, please refer to this link. An extended version of the VIPl-HR database (VIPL-HR-V2) can be accessed using this link.. For the OBF database, please contact Xiaobai Li for more information.

Data Processing

The MSTmap generation procedure is based on Matlab. Please see the MSTmap_generation folder for more information. Both the SeetaFaceEngine (81 landmarks) and OpenFace (68 landmarks) facial landmarks detection engines are supported.


This code is only an toy example of the training procedure. All the network structures and losses are provided. You need to adjust the Dataloader and training and test functions based on your own data.

If you use this work, please cite our paper

title={Video-based Remote Physiological Measurement via Cross-verified Feature Disentangling.},
author={Niu, Xuesong and Yu, Zitong and Han, Hu and Li, Xiaobai and Shan, Shiguang and Zhao, Guoying},
booktitle= {European Conference on Computer Vision (ECCV)},