Remote Photoplethysmography with Sparse and Efficient Deep Neural Networks

This research introduces a novel technique for optimizing remote Photoplethysmography (rPPG) networks through pruning, aiming to create compact yet robust models that operate effectively even with minimal training data. The method emphasizes reducing network complexity without compromising accuracy, specifically tailored for real-time use cases. Experimental findings illustrate that pruned networks achieve comparable performance to their larger counterparts, underscoring their viability for deployment in environments with limited resources.