Promoting Generalization in Cross-Dataset Remote Photoplethysmography
This addresses generalization issues in remote heart rate monitoring for healthcare applications, but it is incremental as it builds on existing deep learning methods with specific augmentations.
The paper tackled the problem of deep learning models in remote photoplethysmography learning dataset-specific biases, and by developing augmentations to expand heart rate variability during training, they reduced mean absolute error from over 13 to below 3 beats per minute in cross-dataset tests.
Remote Photoplethysmography (rPPG), or the remote monitoring of a subject's heart rate using a camera, has seen a shift from handcrafted techniques to deep learning models. While current solutions offer substantial performance gains, we show that these models tend to learn a bias to pulse wave features inherent to the training dataset. We develop augmentations to mitigate this learned bias by expanding both the range and variability of heart rates that the model sees while training, resulting in improved model convergence when training and cross-dataset generalization at test time. Through a 3-way cross dataset analysis we demonstrate a reduction in mean absolute error from over 13 beats per minute to below 3 beats per minute. We compare our method with other recent rPPG systems, finding similar performance under a variety of evaluation parameters.