I've spent the last couple of years working on a new rPPG model, and I just published this blog post (which links to the full technical paper on arXiv) about the 2.0 version of my VitalLens API.
The main breakthrough is that the new model is now accurate enough to robustly measure Heart Rate Variability (HRV) metrics (SDNN, RMSSD) from a webcam feed, not just heart rate.
The blog post covers the new model architecture, the 1,400+ person training dataset, and performance benchmarks.
I'm here to answer any questions about the model, the tech, or the challenges of building a bootstrapped API.
Thanks for checking it out!
guzik•3mo ago
Curious if you've done any comparisons against actual hardware (chest-based or contact PPG), or if validation is mostly dataset based for now.
prouast•2mo ago
I have written up a technical report on the validation here: https://arxiv.org/abs/2510.27028 This validation is based on several datasets, which all have gold standard ground truth from contact PPG and RESP from chest straps and/or impedance pneumography (from ECG)