Uncertainty-quantified Pulse Signal Recovery from Facial Video using Regularized Stochastic Interpolants

TMLR Paper5323 Authors

Abstract:

Imaging Photoplethysmography (iPPG), an optical procedure which recovers a human’s blood volume pulse (BVP) waveform using pixel readout from a camera, is an exciting research field with many researchers performing clinical studies of iPPG algorithms. While current algorithms to solve the iPPG task have shown outstanding performance on benchmark datasets, no state-of-the art algorithms, to the best of our knowledge, performs test-time sampling of solution space, precluding an uncertainty analysis that is critical for clinical applications. We address this deficiency though a new paradigm named Regularized Interpolants with Stochasticity for iPPG (RIS-iPPG). Modeling iPPG recovery as an inverse problem, we build probability paths that evolve the camera pixel distribution to the ground-truth signal distribution by predicting the instantaneous flow and score vectors; and at test-time, we sample the posterior distribution of the correct BVP waveform given the camera pixel intensity measurements by solving a stochastic differential equation. Given that physiological changes are slowly varying, we show that iPPG recovery can be improved through regularization that maximizes the correlation between the residual flow vector predictions of two adjacent time windows. Experimental results on three datasets show that RIS-iPPG provides superior reconstruction quality and uncertainty estimates of the reconstruction, a critical tool for the widespread adoption of iPPG algorithms in clinical and
consumer settings.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

We have made the following changes since our submission on July 7, 2025.

* Updated the text Section 1 (Introduction) to be clearer
* Clarified Ambiguity in Section 3 (Background: Stochastic Interpolants)
* Provided more evidence for our preliminary investigations (Section 4.2)
* Re-wrote significant portions of Section 4.3, 4.4, and 4.5 to better explain our rationale behind stochastic interpolants
* Included new uncertainty quantification metrics (Section 5.2b)
* Generated new uncertainty quantification results (Tables 4 and 5)
* Plotted calibration curves on the PURE and UBFC-rPPG datasets (Figure 5)
* Described the inadequacy of previous methods as an inspiration for our method (Appendx A.3.1)
* Compared against other posterior sampling methods (Appendix A.4)
* Derived the denoiser used in RIS-iPPG (Section A.5)
* Described the intuition behind the RCL loss, including a short proof (Appendix A.6)
* Derived the sampling SDE (Appendix A.7)
* Enumerated the Stability, Data Requirements, and Approximation Quality (Appendix A.8)
* Plotted training cuves (Figure 7)
* Described model performance on protected attributes in Appendix Section A.9.4 (and Figure 9)

Assigned Action Editor: ~Marcus_A_Brubaker1

Submission Number: 5323