RPM-Distill: Physiology-guided Adaptive Cross-modal Distillation for Robust Remote Physiological Measurement

2026-06-26

AI Summary

This work addresses the limited robustness of existing video-based remote physiological measurement methods under challenging conditions such as illumination variations, skin tone differences, and motion artifacts, as well as the practical deployment difficulties of radio-frequency radar despite its stable performance. To bridge this gap, the authors propose a physiology-guided cross-modal distillation framework that leverages synchronized radar signals during training but requires only video input at inference time. The key innovation lies in exploiting, for the first time, the consistency of physiological rhythms between radar and video modalities in the frequency domain for knowledge distillation, complemented by a sample-level adaptive gating mechanism to prevent negative transfer. Integrating spectral matching loss in the frequency domain, structured physiological spectral modeling, and meta bi-level optimization, the method achieves an 81% reduction in mean absolute error and a 21% improvement in correlation over single-modality baselines in both challenging scenarios and cross-dataset evaluations.

Abstract

Video-based remote physiological measurement (RPM) is highly accessible but remains fragile under varying illumination, skin tones, and motion. Radio frequency (RF) radar is largely invariant to illumination and appearance, providing complementary cardio-respiratory micro-motion cues; however, requiring radar at inference is often impractical due to its limited ubiquity and deployment overhead. We propose RPM-Distill, a physiology-guided cross-modal distillation framework that leverages synchronized radar only during training while retaining video-only inference. Our key observation is that although RGB and RF waveforms differ in sensing physics and time-domain morphology, they share similar latent periodic rhythm in the frequency domain. We thus distill physiology-structured spectral evidence to improve robustness, via losses that (i) anchor the fundamental peak, (ii) match the off-peak background distribution, and (iii) preserve spectral morphology and sharpness. To avoid negative transfer under sample-level teacher quality and alignment uncertainty, a spectral policy network predicts sample-level distillation gates and component weights from the student--teacher spectral relation map, learned with a meta bilevel objective on a small labeled validation split. Through extensive experiments in challenging conditions and cross-dataset settings, RPM-Distill brings 81\% MAE and 21\% correlation improvement over unimodal baselines. Code is at https://github.com/WJULYW/RPM-Distill.

Problem

• remote physiological measurement
• cross-modal distillation
• robustness
• video-based sensing
• physiological signal estimation

Innovation

• cross-modal distillation
• physiology-guided learning
• remote physiological measurement
• spectral knowledge transfer
• adaptive gating