We are thrilled to announce the official launch of VitalLens 2.0, the next generation of our remote photoplethysmography (rPPG) model and the new engine powering the VitalLens API.
This isn't just an incremental update; it's a significant leap forward. VitalLens 2.0 achieves state-of-the-art accuracy and, for the first time, enables the robust estimation of Heart Rate Variability (HRV) metrics from face video.
Best of all: VitalLens 2.0 is now available as a free, automatic upgrade for all paid API plans.
From Rates to High-Fidelity Waveforms
Our original model, VitalLens 1.0, was designed to provide accurate heart rate (HR) and respiratory rate (RR) estimations. But accurately measuring HRV, the precise, sub-second variations between heartbeats, is a much harder challenge. It requires moving from simple rate estimation (finding a dominant frequency) to true high-fidelity waveform reconstruction.
VitalLens 2.0 was built to do exactly that. The new model's ability to reconstruct the physiological pulse waveform with high temporal accuracy is what unlocks reliable HRV analysis.
How We Achieved This
This new state-of-the-art (SOTA) performance was made possible by two key developments over the last two years:
- A Massive, Diverse Training Set: We expanded our training data to 1,413 unique individuals, combining our in-house data with public datasets like Vital Videos. This set was meticulously curated and includes a wide diversity of skin tones, lighting conditions, and real-world motion.
- A New Model Architecture: We developed a new end-to-end deep learning architecture based on an EfficientNet backbone. It uses novel temporal-attentive mechanisms to focus on and preserve the subtle, high-frequency details of the pulse, allowing for the precise detection of systolic peaks.
On our new combined test set of 422 unique individuals, VitalLens 2.0 achieves a new SOTA Mean Absolute Error (MAE) of 1.57 bpm for HR, 1.08 bpm for RR, and just 10.18 ms for HRV-SDNN.
What's New For Your Plan
We are rolling out this new technology to all our users today. Here’s how your plan is being upgraded:
-
📈 All Paid Plans: You have been automatically upgraded to the full VitalLens 2.0 model. You can now access HR, RR, and new HRV metrics (SDNN, RMSSD, LF/HF) at no additional cost.
-
✅ Free Plan: You are being upgraded to VitalLens 1.1. This is our original VitalLens 1.0 architecture that has been re-trained on our new, massive dataset. This provides a significant boost in accuracy and robustness for HR and RR estimation, but does not include HRV support.
How to Get the Update
To access the new models, you just need to update your client library to the latest version. Our clients automatically handle model resolution, so you'll always be using the best model available to your plan.
Python Users:
Update your vitallens package using pip: pip install --upgrade vitallens prpy
JavaScript Users:
Ensure your <script> tag is pointing to version 0.2.3 or greater. If you are using a latest tag, you are already set.
Node.js Users:
Update your vitallens package using npm: npm update vitallens.
Direct API Users: You are already set.
Get Started with VitalLens 2.0
- New to VitalLens? Sign up for a free account to start building with our API.
- Already have an account? Upgrade to a paid plan to get immediate access to VitalLens 2.0 and robust HRV estimation.
The Technical Deep Dive
We've published a technical report detailing the VitalLens 2.0 architecture, our new test set, and a comprehensive performance benchmark against other SOTA models. For anyone interested in the data, methodology, and robustness analysis (including performance on motion and across skin types), we invite you to read the full paper.
Read the full paper: VitalLens 2.0: High-Fidelity rPPG for Heart Rate Variability Estimation from Face Video
We're incredibly proud of this release and can't wait to see what you build with it. As always, please contact us with any questions or feedback.
— The Rouast Labs Team