on Dec 7, 2007
This is amazing. I almost didn’t link it because you have to visit the site to see it. They really need to think about making that into a YouTube video.
The demo on that site shows a woman making various expressions and faces for the camera, while a computer-generated face does an exceptional job of imitating her. The actual geometry of the face isn’t all that impressive. (It looks very plastic and lifeless. Haven’t these guys heard of texture maps? Then again, maybe the model is made this way on purpose, so that you focus on the motion and not the shape of the face.) The animation goes a long way to bringing the face to life, although I can’t help but wonder how it would look with a more convincing 3d model.
The obvious application here (and what they seem to be selling) is a system for making complex facial animations with minimal work. Doing this stuff by hand is a chore, and tends to look very mechanical. (I’ve dabbled with it. It sucks, and it takes a huge investment of time and skill to get rid of the mechanical feel.) Doing it “right” involves an actor putting on special makeup (usually dots all over their face for the computer to track) or other tricks to give the computer well-defined points of reference. But the system shown here seems to let anyone sit down in front of the camera and start mugging without any fancy setup.
I’m willing to bet it isn’t realtime, even though the demo movie seems designed to allow you to think so. I’m guessing you capture the video and then the software needs a jolly good think before you can see the results in action. I’m also confident that it would have trouble if you spoke. I doubt it attempts to track anything inside the mouth. Near the end of the demo the actress licks her lips, and the avatar doesn’t. It’s possible that the avatar doesn’t have a tongue, but I think it’s more likely that the software can’t quite comprehend tongues yet. This is a shame, since if that worked this would be the ultimate way to make lip synch look good. Just get a video of your voice actors as they do their thing. Most voice actors already emote visually when they do their thing, so the extra data would be “free”.
At any rate: It’s fun to watch the animation do its thing.
Shamus Young is an old-school OpenGL programmer, author, and composer. He runs this site and if anything is broken you should probably blame him.