Today’s nightmare fodder: the disembodied faces of nine different politicians and celebrities—including Barack Obama—all in sync with the expressive quirks of former president George W. Bush. It’s as though Bush put on nine different photorealistic masks, and the effect is surreal. (The segment in question begins at 1:30 in the video.)
This video was created by a team of researchers from the University of Washington who wanted to answer the question “What makes Tom Hanks look like Tom Hanks?” To do so, they developed machine-learning algorithms that mine the Internet for images of a person and use them to create a three-dimensional reconstruction of that person’s face that captures not only features but also unique expressions and mannerisms.
They can sync one face’s “persona” to one of the re-created faces, as with the example mapping Bush’s mannerisms onto Obama’s face. But their software works in the other direction as well: to map the “performance” of one person onto another’s face while preserving the second face’s persona. As explained by coauthor and professor of computer science Steven Seitz in a press release: “How do you map one person’s performance onto someone else’s face without losing their identity? […] We’ve shown you can have George Bush’s expressions and mouth and movements, but it still looks like George Clooney.”
The team has high hopes for using the technology. What if you could build three-dimensional reconstructions of deceased loved ones to use with the incoming wave of virtual-reality technology; reconstructions that capture the gestalt of the person in question?
To me, they have a ways to go—those videos are uncanny but not quite convincing. Yet it’s an undeniably fascinating proof-of-concept. The press release makes no mention of potential applications in history or archaeology—but wouldn’t it be amazing if we could use the historical record not just to create a three-dimensional image of Abraham Lincoln but one that also captured some of his persona as well?
The team will present its latest findings at the International Conference on Computer Vision in Chile on December 16.