Problem statement

The task of in-the-wild eye contact detection for autonomous agents is under-estimated in the Deep Learning literature today. To the best of our knowledge, there are very few studies about this task and there is a lack of labeled dataset for the latest. Eye-contact detection in the context of autonomous driving is extremely challenging and useful. The autonomous agent should change its behavior whenever there is a pedestrian looking at it, or not. This applies also for social robots, where the interaction with the latest and the social environment is extremely important. We come up with several contributions, including the release in the following weeks of a new benchmark for this task.

Implementation & paper

A detailed presentation of the method, as well as the open source code is available here.