Young-gaze: an appearance-based gaze estimation solution for adolescents

According to the World Health Organization survey, the global incidence of adolescent mental illness is 28%, while the disease detection rate is only 24.6%. Many existing works use complex eye-tracking devices to study adolescent autism, depression, and other mental illness. In this paper, we propose a gaze estimation method to replace eye-tracking devices. Appearance-based methods with deep learning can predict the point of gaze by using a monocular camera, which requires a large number of samples to learn. However, the samples collected in publicly available gaze estimation datasets are mainly adults and not adolescents. To address the above issue, our work makes two contributions. First, we collected images from 107 adolescents aged 10–14 years by laptops under uncontrolled conditions to create the Young-Gaze dataset. Second, we propose a Multi-scale Feature Fusion-based Calibration Network (MFFC-Net) to deeply fuse the eye-face features for gaze estimation. The proposed MFFC-Net achieves the better performance on Young-Gaze and other public datasets.