Gaze is an externally observable indicator of human visual attention, and thus, recording the gaze position can help to solve many problems. Existing gaze estimation models typically utilize separate neural network branches to process data streams from both eyes and the face, failing to fully exploit their feature correlations. This study presents a gaze estimation network that integrates multi-head attention mechanisms, fusion, and interaction strategies to fuse facial features with eye features, as well as features from both eyes, separately. Specifically, multi-head attention and channel attention are used to fuse features from both eyes, and a face and eye interaction module is designed to highlight the most important facial features guided by the eye features; in addition, the channel attention in the Convolutional Block Attention Module (CBAM) is replaced with minimum pooling instead of maximum pooling, and a shortcut connection is added to enhance the network’s attention to eye region details. Comparative experiments on three public datasets—Gaze360, MPIIFaceGaze, and EYEDIAP—validate the superiority of the proposed method.
Recently, with the widespread application of deep learning networks, appearance-based gaze estimation has made breakthrough progress. However, most methods focus on feature extraction from the facial region while neglecting the critical role of the eye region in gaze estimation, leading to insufficient eye detail representation. To address this issue, this paper proposes a multi-stream multi-input network architecture (MSMI-Net) based on appearance. The model consists of two independent streams designed to extract high-dimensional eye features and low-dimensional features, integrating both eye and facial information. A parallel channel and spatial attention mechanism is employed to fuse low-dimensional eye and facial features, while an adaptive weight adjustment mechanism (AWAM) dynamically determines the contribution ratio of eye and facial features. The concatenated high-dimensional and fused low-dimensional features are processed through fully connected layers to predict the final gaze direction. Extensive experiments on the EYEDIAP, MPIIFaceGaze, and Gaze360 datasets validate the superiority of the proposed method.