DMAGaze: Gaze Estimation Based on Feature Disentanglement and Multi-Scale Attention

Gaze estimation, which predicts gaze direction, commonly faces the challenge of interference from complex gaze-irrelevant information in face images. In this work, we propose DMAGaze, a novel gaze estimation framework that exploits information from facial images in three aspects: gaze-relevant global features (disentangled from facial image), local eye features (extracted from cropped eye patch), and head pose estimation features, to improve overall performance. Furthermore, we introduce a new cascaded attention module named Multi-Scale Global Local Attention Module (MS-GLAM). Through a customized cascaded attention structure, it e ffectively focuses on global and local information at multiple scales, further enhancing the information from the Disentangler. Finally, the global gaze-relevant features disentangled by the upper face branch, combined with head pose and local eye features, are passed through the detection head for high-precision gaze estimation. Our proposed DMAGaze has been extensively validated on two mainstream public datasets, achieving state-of-the-art performance. Keywords: gaze estimation, feature disentanglement, Gaussian similarity, multi-scale attention1. Introduction Gaze estimation, the task of predicting gaze direction, crucial for measuring human attention, is widely applied in areas like saliency detection[1, 2], virtual reality[3], driver distraction monitoring[4], human-computer interaction[5] and autism diagnosis[6]. Recently, gaze estimation has shifted from model-based methods to appearance-based methods.