GESCAM : A Dataset and Method on Gaze Estimation for Classroom Attention Measurement

Athul M. Mathew, Arshad Ali Khan, Thariq Khalid, Riad Souissi; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 636-645

Human gaze provides crucial insights into individual attention during social or educational interactions. Attention systems often rely on head and facial features to predict gaze direction but reliable gaze target detection (GTD) requires rich contextual cues. These cues inform the system about an individual's position within a scene and the surrounding objects they might be interacting with. Our paper proposes attention measurement using GTD in educational classrooms leveraging a synthetic dataset called GESCAM (Gaze Estimation based Synthetic Classroom Attention Measurement). This dataset was meticulously generated using 3D modelling animation simulation and rendering techniques comprising 60000 images with 650000 instances of individuals (students teachers) engaged in various activities including looking at blackboard notebooks mobile phones etc. Our novel network trained on GESCAM proficiently identifies gaze fixations within complex classroom scenes offering insights into human attention in classrooms across diverse contexts.