Comparing ResNet and Facial Landmarks for Real-time Student Attention Detection

The article discusses two approaches for detecting student attention levels in a classroom: facial landmarks and deep learning (ResNet). It highlights a recent paper that found people focus primarily on the eyes and mouth when recognizing emotions.

Details

The article presents two main approaches for detecting student attention levels in a classroom setting. The first approach is based on facial landmarks, which are specific coordinate points (x, y) that map key features on a face. The standard model uses 68 points, but a recent paper found that people primarily focus on the eyes (especially the left eye) and mouth when recognizing emotions. This led to an innovation of reducing the landmarks to just 24 critical points. The second approach is deep learning using a ResNet model, which can process raw facial images and output emotion classification directly. Both methods have their merits and tradeoffs in terms of resource requirements and performance for real-time deployment in a resource-constrained classroom environment.