ChatPPG.
Join free
Model

TPA-Flow

tensor-product attention + rectified-flow decoding for rPPG

RGBResearch Square preprint
Cite
PreprintCatalog paper date: 2026Source
CitationsView DOI record (opens in a new tab)

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.

A Direct branch produces a stable waveform anchor; a Flow branch (rectified flow) learns a short transport path conditioned on that anchor and a consensus of physiological candidates to fix residual phase and local morphology errors. Physiological Tensor-Product Attention injects a confidence-weighted periodic prior and uses low-rank tensor products across facial regions, cardiac cycles, and feature subspaces.

Signals
RGB
Form factor
camera_rppg
License
CC BY 4.0
Year
2026
Architecture

tensor-product attention plus rectified-flow waveform decoder for rPPG

At a glance
Type
Task model
Modalities
RGB
Form factor
Camera rPPG
Parameters
Not reported
Architecture
tensor-product attention plus rectified-flow waveform decoder for rPPG
Venue
Research Square preprint (not peer reviewed), DOI 10.21203/rs.3.rs-11099271/v1, posted 2026-10-07
Year
2026
arXiv
Not reported
DOI
10.21203/rs.3.rs-11099271/v1 (opens in a new tab)
Code
Not reported
Weights
Not reported
Licence
CC BY 4.0 (opens in a new tab)
Caveats
Not reported
Datasets
Not reported
Added
Not reported
Updated
Oct 27, 2026
TPA-Flow: ChatPPG research summary card

How it works

How TPA-Flow works: Input signals → Model → Downstream tasksInput signals: RGB. Model: TPA-Flow (Transformer).Input signalsRGBRGBModel×NTPA-FlowTransformer
Architecture
tensor-product attention plus rectified-flow waveform decoder for rPPG
Inputs
RGB
Approach
rppg · model · attention · rectified flow · generative decoding · waveform morphology

Datasets

No public datasets linked yet.

Cite

APA

TPA-Flow. (2026). Research Square preprint , DOI 10.21203/rs.3.rs-11099271/v1, posted 2026-10-07. https://doi.org/10.21203/rs.3.rs-11099271/v1

BibTeX

@misc{tpaflow2026tpaflow,
  title         = {{TPA-Flow}},
  year          = {2026},
  howpublished  = {Research Square preprint , DOI 10.21203/rs.3.rs-11099271/v1, posted 2026-10-07},
  doi           = {10.21203/rs.3.rs-11099271/v1}
}

Built from the hub record. Check it against the publisher before submitting.

History

  1. Record updated

Same tasks, data and authors