1 paper · 1 filter
Yuki Kawana, Shintaro Shiba, Quan Kong +1
We propose a novel 3D gaze estimation approach that learns spatial relationships between the subject and objects in the scene, and outputs 3D gaze direction. Our method targets unc…