Sparse and Efficient Deep Neural Networks for remote PPg Applications

This research presents an innovative approach to optimizing remote Photoplethysmography (rPPG) networks by employing pruning techniques. The goal is to create smaller, more efficient models that maintain high levels of performance, even with limited training data. The proposed method focuses on minimizing network complexity while preserving accuracy, making it particularly suitable for real-time applications. The study's experimental results demonstrate that the pruned networks can perform on par with their larger, more complex counterparts. This suggests that these streamlined models are a viable option for deployment in resource-constrained environments, where computational power and memory are limited. By achieving a balance between efficiency and accuracy, this research contributes to the practical application of rPPG technology in various real-world scenarios.