Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.
Frequency-first rPPG. The input passes through a physiological bandpass (out-of-band rejection), then a spectrum-modulation block with adaptive spectral selection that boosts pulse-band components and suppresses in-band noise. Spectral and time-domain features are fused, and a frequency-aware conditional diffusion process reconstructs the waveform.
Physiological Bandpass Filtering → Physiological Spectrum Modulation (real/imag encoding) + adaptive spectral selection → Cross-domain Representation Learning (spectral priors fused with deep time-domain features) → frequency-aware conditional diffusion that reconstructs the rPPG signal
Remote photoplethysmography (rPPG) enables contactless physiological monitoring by capturing subtle skin-color variations from facial videos. However, most existing methods predominantly rely on time-domain modeling, making them vulnerable to motion artifacts and illumination fluctuations, where weak physiological clues are easily overwhelmed by noise. To address these challenges, we propose FreqPhys, a frequency-guided rPPG framework that explicitly leverages physiological frequency priors for robust signal recovery. Specifically, FreqPhys first applies a Physiological Bandpass Filtering module to suppress out-of-band interference, and then performs Physiological Spectrum Modulation together with adaptive spectral selection to emphasize pulse-related frequency components while suppress residual in-band noise. A Cross-domain Representation Learning module further fuses these spectral priors with deep time-domain features to capture informative spatial–temporal dependencies. Finally, a frequency-aware conditional diffusion process progressively reconstructs high-fidelity rPPG signals. Extensive experiments on six benchmarks demonstrate that FreqPhys yields significant improvements over state-of-the-art approaches, particularly under challenging motion conditions. It highlights the importance of explicitly modeling physiological frequency priors. The source code will be released.
Verbatim fromarXiv· arXiv metadata (abstracts are released under CC0)
Datasets linked in the hub; pretraining vs evaluation roles follow the paper.
| Task | Dataset | Metric | Value | Rank |
|---|---|---|---|---|
| Rppg hr | BUAA-MIHR intra | HR MAE ↓ | 1.13 | #2 / 4 |
| Rppg hr | PURE intra | HR MAE ↓ | 0.17 | #2 / 5 |
| Rppg hr | UBFC-rPPG intra | HR MAE ↓ | 0.17 | #4 / 5 |
| Rppg hr | mmpd intra | HR MAE ↓ | 4.2 | #4 / 6 |
| Rppg hr | mr-nirp MR-NIRP-Car intra | HR MAE ↓ | 5.75 | #1 / 1 |
| Rppg hr | pure → mmpd | HR MAE ↓ | 10.11 | #2 / 3 |
| Rppg hr | pure → ubfc-rppg | HR MAE ↓ | 0.43 | #2 / 3 |
| Rppg hr | ubfc-rppg → mmpd | HR MAE ↓ | 8.91 | #1 / 2 |
FreqPhys. (2026). ECCV 2026. https://arxiv.org/abs/2604.00534
@inproceedings{freqphys2026freqphys,
title = {{FreqPhys}},
year = {2026},
booktitle = {ECCV 2026},
eprint = {2604.00534},
archivePrefix = {arXiv}
}Record updated
