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.
StreamPPG predicts an rPPG sample for each incoming frame (causal streaming) instead of buffering a multi-second clip. Training uses consistent privileged learning (CPL): ground-truth BVP is available as privileged input during training and dropped at inference, with a consistency term so train and test behavior stay aligned. Paper: arXiv:2606.23186 (v2 2026-09-28).
dual 2D-CNN encoders (frame + temporal-difference) with channel attention, adaptive temporal modeling module (ATMM) / temporal state-space block, lightweight MLP decoder; trained under consistent privileged learning (CPL) that uses GT rPPG as privileged info at train time only
Remote photoplethysmography (rPPG) estimates the blood volume pulse (BVP) signal from facial videos, enabling contact-free health monitoring. Conventional clip-wise approaches, which use video clips as input, require capturing over one hundred frames before inference, thus introducing several seconds of delay and hindering real-time use. Meanwhile, frame-wise approaches struggle to capture long-range temporal and periodic features of physiological rhythms, and therefore lead to reduced estimation accuracy. To overcome these issues, we propose StreamPPG, a unified architecture that enables low-latency frame-wise physiological signal estimation while achieving competitive accuracy compared with clip-wise approaches. StreamPPG is trained under a consistent privileged learning (CPL) strategy, which leverages ground-truth rPPG signals as privileged information to enhance the model's representation capability. Extensive experiments demonstrate that StreamPPG achieves state-of-the-art accuracy across multiple datasets while maintaining real-time throughput on edge devices.
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 | COHFACE intra | HR MAE ↓ | 0.71 | #2 / 2 |
| Rppg hr | PURE intra | HR MAE ↓ | 0.2 | #5 / 5 |
| Rppg hr | UBFC-rPPG intra | HR MAE ↓ | 0.14 | #2 / 5 |
| Rppg hr | mmpd intra | HR MAE ↓ | 3.39 | #2 / 6 |
| Rppg hr | pure → mmpd | HR MAE ↓ | 10.42 | #3 / 3 |
| Rppg hr | pure → ubfc-rppg | HR MAE ↓ | 0.33 | #1 / 3 |
| Rppg hr | ubfc-rppg → mmpd | HR MAE ↓ | 9.29 | #2 / 2 |
| Rppg hr | ubfc-rppg → pure | HR MAE ↓ | 2.0 | #2 / 2 |
StreamPPG. (2026). arXiv preprint. https://arxiv.org/abs/2606.23186
@misc{streamppg2026streamppg,
title = {{StreamPPG}},
year = {2026},
howpublished = {arXiv preprint},
eprint = {2606.23186},
archivePrefix = {arXiv}
}Record updated
