HSRD (Hierarchical Style-aware Representation Disentangling)

Official code of IEEE JBHI 2024 "Hierarchical Style-Aware Domain Generalization for Remote Physiological Measurement".

Data Prepare

You can refer to link to obtain the processed STMaps. Before that, please get the permission to use the following datasets first: VIPL, V4V, BUAA, UBFC, PURE. After getting STMaps, you can create a new './STMap' folder and put them into it. For the first time running, please adjust the hyperparameter 'reData' to 1, to generate the STMap index.

Pre-trained Model

In this work, we utilized the ResNet18 as the backbone network. You can download it directly from this link. Next, create a new folder './pre_encoder' and put the pth file into it.

Train and Test

Then, you can try to train it with the following command:

python train.py -g $GPU id$ -t 'the target dataset you want to test on'

Citation

@ARTICLE{10371379,
  author={Wang, Jiyao and Lu, Hao and Wang, Ange and Chen, Yingcong and He, Dengbo},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={Hierarchical Style-Aware Domain Generalization for Remote Physiological Measurement}, 
  year={2024},
  volume={28},
  number={3},
  pages={1635-1643},
  keywords={Feature extraction;Videos;Skin;Physiology;Biomedical measurement;Bioinformatics;Training;Adversarial learning;contrastive learning;domain generalization;heart rate estimation;remote photoplethysmography (rPPG)},
  doi={10.1109/JBHI.2023.3346057}}