DG-rPPG

Official code of IEEE TIM 2024 "Generalizable Remote Physiological Measurement via Semantic-Sheltered Alignment and Plausible Style Randomization"

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.

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. For the first time running, please adjust the hyperparameter 'reData' to 1, to generate the STMap index.

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' -tk 'the TopK sample for PAR loss' -si 'standard interval for SSA loss'

Citation

@ARTICLE{dg2024wang,
  author={Wang, Jiyao and Lu, Hao and Han, Hu and Chen, Yingcong and He, Dengbo and Wu, Kaishun},
  journal={IEEE Transactions on Instrumentation and Measurement}, 
  title={Generalizable Remote Physiological Measurement via Semantic-Sheltered Alignment and Plausible Style Randomization}, 
  year={2024}
}