Károly Zsolnai-Fehér explains Face2Face, a new facial reenactment method that works with color data alone and synthesizes mouths instead of copying them. The technology raises concerns about video evidence in legal cases.
Dear fellow scholars, this is Two Minute Papers Witheher. There was a previous episode on a technique where the inputs were a source video of ourselves and a target actor and the output was a video of this target actor with our facial gestures. With such an algorithm, one can edit pre recorded videos in real time, and the current version only needs a consumer webcam to do that. This new version addresses two major shortcomings. One, the previous work relied on depth information, which means that we needed to know how far different parts of the image were from the camera. This newer version only relies on color information and does not need anything beyond that. Whoa. And two previous techniques often resorted to copying the footage from the mouth and adding synthetic proxies for teeth. Not anymore. With this one. I tip my hat to the authors who came up with a vastly improved version of their previous method so quickly, and it is probably needless to say that the ramifications of such an existing technique are far reaching and are hopefully pointed in a positive direction. However, we should bear in mind that from now on we may be one step closer to an era where a video of something happening won't be taken as proper evidence. I wonder how this will affect legal decision making in the future. Thanks for watching and for your generous support, and I'll see you next time.
