News authors: Jitesh Joshi and Youngjun Cho
🚀 Exciting to announce that our work, “FactorizePhys: Matrix Factorization for Multidimensional Attention in Remote Physiological Sensing,” has been accepted and presented at NeurIPS 2024! 🎉
In this work, we introduce the Factorized Self-Attention Module (FSAM) and the FactorizePhys architecture, a cutting-edge approach to pulse rate estimation from video. By leveraging nonnegative matrix factorization, we developed a novel multidimensional attention mechanism that computes spatial, temporal, and channel attention jointly—setting a new benchmark in remote photoplethysmography (rPPG).
Highlights:
✅ Multidimensional attention enabling superior spatial-temporal feature extraction.
✅ Outstanding cross-dataset generalization demonstrated across benchmarks.
✅ Open-source code to inspire and support future research.
We’re excited about the potential of this research in healthcare applications, from stress and mental health monitoring to driver drowsiness detection and beyond.
📰 Read our paper: https://openreview.net/pdf?id=qrfp4eeZ47
📰 Quick glance at our tech (slides): https://neurips.cc/media/neurips-2024/Slides/93470_NmYSJaO.pdf
💻 Explore our code: https://github.com/PhysiologicAILab/FactorizePhys