LQ-rPPG: A Label-Quantized Coarse-to-Fine Learning Framework for Remote Physiological Measurement
Researchers have developed LQ-rPPG, a novel framework designed to improve the accuracy of remote physiological measurements from facial videos. This method addresses the issue of noisy and inconsistent training labels in existing deep learning models by transforming continuous physiological signals into quantized pseudo-labels. The framework then employs a coarse-to-fine estimation model that progressively refines the signals using hierarchical supervision, leading to more robust and generalizable rPPG estimation. LQ-rPPG not only enhances performance but also significantly reduces model parameters and computational load while increasing throughput. AI
IMPACT Enhances accuracy and efficiency in remote health monitoring by improving physiological signal extraction from video.