CanonicalPhys: Pose-Robust Remote Photoplethysmography via Canonical-Space Priors

2026-07-17

AI Summary

This study addresses the significant performance degradation of existing deep rPPG methods under head pose variations, which stems from unstable pixel-to-anatomy mapping in the image coordinate system that disrupts the physical priors essential for rPPG. The work reframes pose variation as a structural coordinate perturbation rather than a data augmentation challenge and introduces a parameter-free canonical space modeling approach. It employs a differentiable four-point homography to align facial landmarks into a canonical space, where rPPG priors are reconstructed using Lambertian reflectance weighting, cross-regional temporal consistency loss, and window-based POS knowledge distillation. Experiments demonstrate substantial mitigation of pose-induced degradation on MMPD: MAE deterioration under large yaw angles is reduced from 1.60× to 1.33×, and from 1.32× to 1.07× under moderate yaw, with cross-dataset MAE improvements of up to 32%.

Abstract

Deep remote photoplethysmography (rPPG) attains sub-bpm heart-rate error on frontal, stationary faces yet degrades sharply under head pose: on MMPD, the state-of-the-art FactorizePhys backbone's MAE grows $1.60\times$ from frontal ($|\text{yaw}|{<}15^\circ$) to large-yaw ($|\text{yaw}|{\geq}45^\circ$) frames. We argue that pose is a \emph{coordinate-structural} nuisance rather than a data-augmentation problem: in image coordinates the same pixel maps to different anatomy at different poses, blocking three priors otherwise natural for rPPG, namely the dichromatic reflection model, pulse-phase invariance across skin regions, and the POS/CHROM chromaticity projection, each of which presumes a stable anatomy-to-pixel mapping. We introduce \textbf{CanonicalPhys}, which prepends a differentiable four-point homography that fixes four facial anchors at canonical positions; in this canonical frame the three priors become expressible as a per-pixel Lambertian weight, a cross-ROI temporal consistency loss, and knowledge distillation from windowed POS, none of which adds trainable parameters over the backbone. At an identical parameter count, CanonicalPhys reduces MMPD's frontal-to-large-yaw MAE degradation from $1.60\times$ to $1.33\times$ and flattens the mild-yaw bin from $1.32\times$ to $1.07\times$ (across CanonicalPhys variants), with matched cross-dataset MAE reductions of up to $32\%$ on pose-rich targets. Code: https://github.com/infraface/CanonicalPhys

Problem

• remote photoplethysmography
• head pose variation
• anatomy-to-pixel mapping
• pose robustness
• physiological priors

Innovation

• canonical-space
• pose-robust rPPG
• homography alignment
• physiological priors
• knowledge distillation