VitalVideo, UBFC-rPPG, PURE, MMPD

Dataset Overview

Dataset Names and Descriptions

• VitalVideo: Contains 900 participants covering six skin tones; currently the largest real‑world rPPG dataset.
• UBFC‑rPPG: Used for remote photoplethysmography research.
• PURE: Dataset for video‑based pulse rate detection.
• MMPD: Multi‑Domain Mobile Physiological Data dataset.

Dataset Organization

• MMPD: Organized with one folder per participant, containing multiple .mat files.
• UBFC‑rPPG: Organized with one folder per participant, containing video files and ground‑truth text files.
• PURE: Complex folder structure; each participant folder includes videos and JSON files.
• VitalVideo: Multiple sub‑folders, each containing videos and JSON files.

Usage Guidelines

• When using these datasets, follow the specified folder organization.
• Cite the corresponding papers when employing the datasets in deep‑learning models.

Experiments and Benchmarks

Experiment Overview

• Cross‑dataset experiments were conducted, focusing mainly on the VitalVideo dataset.
• Six unsupervised methods and three state‑of‑the‑art supervised models (TSCAN, PhysNet, PhysFormer) were evaluated.

Results Summary

• Mean Absolute Error (MAE) performance of six unsupervised methods on vv100 and vvAll.
• MAE performance when training on vv100 and testing on PURE, UBFC‑rPPG, MMPD.
• MAE performance when training on PURE, UBFC‑rPPG, MMPD and testing on vv100.

Neural Network Training Examples

• Detailed steps for training on VitalVideo and testing on MMPD.
• Steps for using a pretrained model to train on VitalVideo and test on PURE.

Citation Information

• When using these datasets, cite the relevant publications.