Awesome

iBVP Dataset

This repo is associated with our following work, and provides instructions on how to access the dataset and the code.

Joshi, Jitesh, and Youngjun Cho. 2024. "iBVP Dataset: RGB-Thermal rPPG Dataset with High Resolution Signal Quality Labels" Electronics 13, no. 7: 1334. https://doi.org/10.3390/electronics13071334

About iBVP Dataset

The iBVP dataset is a collection of synchronized RGB and thermal infrared videos with PPG ground-truth signals acquired from an ear. The PPG signals are marked with manual signal quality labels, as well as with the SQA-PhysMD model trained and validated to conduct dense (per-sample) signal-quality assessment. The data acquisition protocol was designed to inducing real-world variations in psycho-physiological states, as well as head movement. Each participant experienced four conditions, including (a) rhythmic slow breathing and rest – “A,” (b) an easy math task – “B,” (c) a difficult math task – “C,” and (d) a guided head movement task – “D.” RGB and thermal cameras were positioned in front of the participant at around a 1 m distance. A webcam (Logitech BRIO 4K UHD) was used to capture RGB video frames with 640 × 480 resolution, while thermal infrared frames were captured using thermal camera (A65SC, FLIR system) having 640 × 512 resolution. Frame-rate was set to at 30 FPS for both RGB and thermal acquisition. With 124 sessions, each lasting 3 minutes, the dataset comprises 372 minutes (about 6 hours) of RGB–Thermal video recordings.

Requesting iBVP Dataset

We have released the iBVP dataset only for academic research purposes. The dataset can be requested from the authors by submitting a signed copy of the end-user's licence agreement EULA. Please note that the EULA form is required to be filled by academic supervisors. After submitting the signed EULA to the email addresses mentioned in the EULA, your request will be reviewed and on acceptance, you will receive a link to download the dataset.

In your email, please include following along with the signed EULA:

• Some words on your research and how the database would be used.
• How you heard of the database (colleague, papers, etc.)

Dataset Contents

The dataset size (compressed) is ~400 GB. After downloading and extracting the zipped data files, the data needs to be organized as mentioned in the folder structure below:

iBVP_Dataset/
|   |-- p01_a/
|      |-- p01_a_rgb/
|      |-- p01_a_t/
|      |-- p01_a_bvp.csv
|   |-- p01_b/
|      |-- p01_b_rgb/
|      |-- p01_b_t/
|      |-- p01_b_bvp.csv
|   ...
|   |-- pii_x/
|      |-- pii_x_rgb/
|      |-- pii_x_t/
|      |-- pii_x_bvp.csv

• pii_x indicates following:
  • ii: Participant ID
  • x: Experimental condition (one out of "a", "b", "c" and "d" as described above).
• pii_x_rgb: Directory consisting of RGB frames (.bmp)
• pii_x_t: Directory consisting of Thermal frames (.raw)
• pii_x_bvp: .csv file with following columns:
  • BVP: Filtered PPG signals, downsampled at 30 FPS to match with the RGB and thermal video frames.
  • SQPhysMD: Signal quality labels generated by our traine SQA_PhysMD model.
  • SQ1: Manually annotated signal quality labels
  • SQ2: Manually annotated signal quality labels
  • Perfusion_Value: Perfusion index computed from the raw PPG signals.

Please note: The data of few participants who provided limited consent is kept in a separate folder, named "Confidential_No-media-use". The participant IDs with limited consent include following: p08, p10, p13, p16, p29, p31, and p33. Though for training and/or evaluating rPPG methods, this data will have to be moved to the main dataset folder (i.e. iBVP_Dataset folder as described above), please keep this data confidential and extremely secured.

Code associated with the iBVP dataset

The code to support the usage of this dataset (e.g. dataloaders for RGB and Thermal video frames) is developed by Youngjun's research group on computational physiology and intelligence at UCL GDIH - WHO Collaborating centre for AT, UCL Interaction Centre, and UCL Computer Science. This code is integrated with the rPPG-Toolbox .

iBVPNet Model

Implementation of the iBVPNet model can be found here.

MACC Evaluation Metrics

The implementation of Maximum Amplitude of Cross-correlation (MACC) can be found here. The illustrative infer configs that enables computing MACC metrics is demonstrated here.

SQA-PhysMD

For the signal quality assessment module (SQA_PhysMD), as proposed in this paper, the model, inference code and the checkpoint can be found here. All .csv files of the iBVP dataset can be copied to this folder to run inference using SQA-PhysMD model. Any raw PPG signals stored in .csv file format can also be used by appropriately changing the data->total_duration_sec and data->window_len_sec values in the config file SQAPhysMD.json. Inference code can be executed with a terminal command as illustrated below:

python SQA_PhysMD/test_SQAPhysMD.py --config SQA_PhysMD/configs/SQAPhysMD.json --datadir data/ppg_sq --savedir data/ppg_sq_out --preprocess 1

SQA_PhysMD is now also integrated with our PhysioKit repository for real time signal quality assessment of PPG signals. The upsampled version (64 samples per second) in .npy format can be further provided upon request for benchmarking with existing signal quality assessment methods as compared in this paper.

Additional Support or Reporting Issues with the Library

For suggestions as well as discussing ideas, please use the discussion space. Bugs or problems faced while using the iBVP dataset can be reported to the Issues section.

Citations

If you find our paper or this dataset useful for your research, please cite our following works.

@article{joshi2024ibvp,
    title={iBVP Dataset: RGB-Thermal rPPG Dataset with High Resolution Signal Quality Labels},
    author={Joshi, Jitesh and Cho, Youngjun},
    journal={Electronics},
    publisher={MDPI},
    volume={13},
    year={2024},
    number={7},
    article-number={1334},
    url={https://www.mdpi.com/2079-9292/13/7/1334},
    issn={2079-9292},
    doi={10.3390/electronics13071334}
}

@article{joshi2023physiokit,
    title={PhysioKit: An Open-Source, Low-Cost Physiological Computing Toolkit for Single-and Multi-User Studies},
    author={Joshi, Jitesh and Wang, Katherine and Cho, Youngjun},
    journal={Sensors},
    publisher={MDPI},
    volume={23},
    number={19},
    article-number={8244},
    year={2023},
    url={https://www.mdpi.com/1424-8220/23/19/8244},
    issn={1424-8220},
    doi={10.3390/s23198244}
}

Additionally, as we build upon rPPG-Toolbox, we would like to sincerely thank and acknowledge the authors of the rPPG-Toolbox. Therefore, we also request the users of the iBVP dataset to cite the rPPG-Toolbox paper:

@inproceedings{liu2024rppg,
    author = {Liu, Xin and Narayanswamy, Girish and Paruchuri, Akshay and Zhang, Xiaoyu and Tang, Jiankai and Zhang, Yuzhe and Sengupta, Roni and Patel, Shwetak and Wang, Yuntao and McDuff, Daniel},
    booktitle = {Advances in Neural Information Processing Systems},
    editor = {A. Oh and T. Neumann and A. Globerson and K. Saenko and M. Hardt and S. Levine},
    pages = {68485--68510},
    publisher = {Curran Associates, Inc.},
    title = {rPPG-Toolbox: Deep Remote PPG Toolbox},
    url = {https://proceedings.neurips.cc/paper_files/paper/2023/file/d7d0d548a6317407e02230f15ce75817-Paper-Datasets_and_Benchmarks.pdf},
    volume = {36},
    year = {2023}
}