Daily CVPR - Wednesday

The greatest moment in yesterday’s presentations was certainly the very brave live video demo that Matthias Nießner and the Face2Face team made at the Computational Photography and Faces orals (actually all presenters there were very impressive). The Face2Face research reenacted a monocular target video sequence (e.g. YouTube) based on the expressions of a source actor. The input to the method is two monocular video sequences: the first one features the target actor to be re-enacted and a second one features the source actor (e.g captured live with a commodity webcam). The output is a new video in which the facial expressions of the target actor are animated by the source actor in a photo-realistic fashion. Yesterday’s live demo source actor offered very funny face expressions to Russian President Vladimir Putin . The effect was so exhilarating that we were not even able to take a picture for our readers. I am therefore indebted to XING ZHANG (Mage Inc. /SUNY- Binghamton), for kindly allowing me to publish the one he took, reproduced below. You will not be surprised to learn that this work was already known to Computer Vision News, by which the Face2Face model was reviewed in the May issue. Read the full review here (pages 16 to 18). 4 Highlights CVPR Daily: Wednesday How to offer one’s own face expressions live to a very compliant V. Putin