Gaze Following Based on Attention Mechanism

We humans have the ability to follow others’ gaze and estimate their gaze targets, which is called gaze following. In social communication, we are constantly doing gaze following to gain more information. With the development of society, gaze following has gained more significant applications, such as offline retail and assistive healthcare. To automate the process of gaze following, people hope that computers can also acquire this ability. However, there are relatively few relevant studies in the field of computer vision, and many studies have failed to focus on image features that are helpful for gaze following. In this paper, we propose a gaze following method based on attention mechanism, which enables the gaze direction estimation network to focus on more effective spatial features and the saliency estimation network to focus on more effective channel features. By concatenating the two networks, we ultimately obtain the predicted heatmap of the gaze point. The experimental results show that our method can effectively improve the performance of gaze following.