Appearance Based Gaze Estimation

Appearance-based gaze estimation aims to predict where a person is looking using only an image of their face, seeking to overcome limitations of traditional methods. Current research focuses on improving generalization across individuals and domains, employing techniques like federated learning for privacy, adversarial training to reduce overfitting to specific appearances, and data augmentation strategies including novel view synthesis and super-resolution to enhance model robustness and accuracy. These advancements are crucial for deploying gaze estimation in real-world applications such as human-computer interaction, assistive robotics, and marketing research, where diverse conditions and user variability are common.