FaceEat: Facial and Eating Activities Recognition with Inertial and Mechanomyography Fusion using a Glasses-Based Design for Real-Time and on-the-Edge Inference.