Deep Learning with Pseudo-Labeling Enables Contactless Heart-Rate Estimation from Video
Researchers describe a deep learning method for remote photoplethysmography (rPPG), which estimates heart rate from ordinary video without any physical contact, a capability aimed at telemedicine use. The approach relies on pseudo-labeling to lessen the need for manually annotated training data, and the authors report state-of-the-art results for contactless heart-rate estimation.