Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PhysMamba: State Space Duality Model for Remote Physiological Measurement

Zhixin Yan, Yan Zhong, Hongbin Xu, Wenjun Zhang, Shangru Yi, Lin Shu, Wenxiong Kang

TL;DR

PhysMamba tackles non-contact physiological measurement (rPPG) under motion and lighting variations by introducing a dual-pathway architecture that fuses State Space Duality with attention mechanisms in a time-frequency interacting framework. Its Multi-Scale Query mechanism enables efficient cross-scale information exchange between self-attention and cross-attention pathways, while Frame Stem, SSSD, and Frequency Domain Feed-Forward enhance temporal and periodic signal representations. Across three public datasets, PhysMamba achieves state-of-the-art intra- and cross-dataset performance with strong generalization and robustness, supporting real-time, non-contact health monitoring in diverse environments. The work provides a scalable foundation for remote health analytics and opens avenues for broad deployment in mobile and telemedicine contexts.

Abstract

Remote Photoplethysmography (rPPG) enables non-contact physiological signal extraction from facial videos, offering applications in psychological state analysis, medical assistance, and anti-face spoofing. However, challenges such as motion artifacts, lighting variations, and noise limit its real-world applicability. To address these issues, we propose PhysMamba, a novel dual-pathway time-frequency interaction model based on Synergistic State Space Duality (SSSD), which for the first time integrates state space models with attention mechanisms in a dual-branch framework. Combined with a Multi-Scale Query (MQ) mechanism, PhysMamba achieves efficient information exchange and enhanced feature representation, ensuring robustness under noisy and dynamic conditions. Experiments on PURE, UBFC-rPPG, and MMPD datasets demonstrate that PhysMamba outperforms state-of-the-art methods, offering superior accuracy and generalization. This work lays a strong foundation for practical applications in non-contact health monitoring, including real-time remote patient care.

PhysMamba: State Space Duality Model for Remote Physiological Measurement

TL;DR

PhysMamba tackles non-contact physiological measurement (rPPG) under motion and lighting variations by introducing a dual-pathway architecture that fuses State Space Duality with attention mechanisms in a time-frequency interacting framework. Its Multi-Scale Query mechanism enables efficient cross-scale information exchange between self-attention and cross-attention pathways, while Frame Stem, SSSD, and Frequency Domain Feed-Forward enhance temporal and periodic signal representations. Across three public datasets, PhysMamba achieves state-of-the-art intra- and cross-dataset performance with strong generalization and robustness, supporting real-time, non-contact health monitoring in diverse environments. The work provides a scalable foundation for remote health analytics and opens avenues for broad deployment in mobile and telemedicine contexts.

Abstract

Remote Photoplethysmography (rPPG) enables non-contact physiological signal extraction from facial videos, offering applications in psychological state analysis, medical assistance, and anti-face spoofing. However, challenges such as motion artifacts, lighting variations, and noise limit its real-world applicability. To address these issues, we propose PhysMamba, a novel dual-pathway time-frequency interaction model based on Synergistic State Space Duality (SSSD), which for the first time integrates state space models with attention mechanisms in a dual-branch framework. Combined with a Multi-Scale Query (MQ) mechanism, PhysMamba achieves efficient information exchange and enhanced feature representation, ensuring robustness under noisy and dynamic conditions. Experiments on PURE, UBFC-rPPG, and MMPD datasets demonstrate that PhysMamba outperforms state-of-the-art methods, offering superior accuracy and generalization. This work lays a strong foundation for practical applications in non-contact health monitoring, including real-time remote patient care.
Paper Structure (19 sections, 8 equations, 5 figures, 4 tables)

This paper contains 19 sections, 8 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. rPPG Estimation
  4. State Space Model
  5. Methodology
  6. Overall Framework
  7. Frame Stem
  8. Multi-Scale Synergistic State Space Duality (SSSD)
  9. Frequency Domain Feed-Forward (FDF)
  10. rPPG Predictor
  11. Experiments
  12. Datasets
  13. Experimental Details
  14. Experimental Evaluation
  15. Intra-Dataset Testing
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Comparison between SSD and SSSD. The traditional SSD combines SSM and attention mechanisms, while our enhanced SSSD introduces Multi-Scale Queries for efficient information interaction between the two pathways.
  • Figure 2: Overall framework diagram of the PhysMamba model. The model comprises the Frame Stem, Multi-Scale Synergistic State Space Duality (SSSD), Frequency Domain Feed-Forward (FDF), and rPPG Predictor, integrated into two pathways: the Self-Attention Pathway and the Cross-Attention Pathway.
  • Figure 3: Comparison of predicted and ground-truth heart rate distributions across UBFC-rPPG, PURE, and MMPD datasets.
  • Figure 4: Visualization of the results from the MMPD dataset. (a) the Bland-Altman plot. (b) the waveform of a sample before and after filtering.
  • Figure 5: Heat map of cross-dataset results: Training on MMPD and testing on PURE (left), and vice versa (right). PhysMamba shows superior generalization compared to other methods.