Abstract:
With the advance of computer vision and deep learning, many problems that were previously thought infeasible are now being solved, whose performance even surpassing human level at certain tasks. But are they really useful enough in practice for understanding real world problems?
In this talk, I share my journey towards answering this question where I investigate the problem of human visual attention. Visual attention is a critically important aspect of human behavior, and reveals rich information about their social, cognitive, and mental states. Although gaze tracking technology has existed for a long time, it carries many inherent physical limitations. I will first show how we can tackle such limitations by formulating the problem as a new computer vision approach. Then, I will present how we can use proposed methods to identify meaningful gaze behaviors, such as eye contact, joint attention and visual search. Lastly, I will share how this new approach enables scalable data analysis and lays the groundwork for scientific discoveries.