Learning-based Gaze Correction for Videoconferencing


During a videoconference the disparity between the face on the screen and the web camera means that a person cannot look simultaneously both into the camera and into the other person’ eyes on her/his screen. Consequently, it is currently impossible for two people to engage into a normal “eye-to-eye” conversation using a conventional laptop and a program such as Skype. Skoltech Computer Vision Group has developed a technology that can address the problem of gaze contact in video conferencing via real-time modification of videostreams in a way that redirects the gazes of participants by a certain angle. In this way, the gaze contact between participants can be reestablished. The technology is purely software-based and does not require any precalibration.