Developers Create Controllable Models of Faces Using a Library of Photos and Video of Celebrities

A team of developers at The University of Washington have developed controllable models of faces using a library of photographs and videos of celebrities including Tom Hanks, Daniel Craig, and Ian McKellen.

The team has published their work in a paper they will present at the International Conference on Computer Vision.

The model is demonstrated in a video that shows the digital masks being altered to match photographs and even sync up with video. A particularly impressive feature of the models is their ability to map on digital mask to the actions of an entirely different person from a video. In one of the examples in the video, a digital mask of President Barack Obama is mapped from a video of former President George W. Bush.

To demonstrate how a single video source could be used to control multiple digital masks, the creators show a source clip of Bush being spoken by nine other computer models.

via The Atlantic