Robustness of non-contact-based heart rate monitoring system

For any neural network to be successful, the most important part of any model is the dataset. Based on the data, model could be biased, underfit or overfit.

Hence, when handling problem statements like non-contact based heart rate monitoring system, the dataset should be diverse enough. Synthetic data creation could help us in achieving that and some works like styleGAN, GANimation and recent work on First Order Motion Model for Image Animation and Animating Arbitrary Objects via Deep Motion Transfer, that generates a video sequence so that an object in a source image is animated according to the motion of a driving video, is a big source to create the synthetic data for training and validation. A sample work is attached below. The project is still in progress and the details would be released soon. Till then do explore the website: here