Show HN: VitalLens 2.0 – A new SOTA rPPG model for HRV estimation from video

[prouast]

Hi HN, I'm Philipp, the solo founder of Rouast Labs.

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]

Congrats guys on your release. I tried new VitalLens side by side with my medical grade tracker (Aidmed One) but the respiration numbers were quite off.

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]

Thanks! Curious under what circumstances the respiration was off - were you standing or sitting down? I can usually see the resp wave go up and down as I breathe while sitting down.

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)