Ok, this seems to be doing the rounds on various news sites, so I'll post it here if anyone hasn't seen it. It looks pretty spectacular, but what they're sorta not telling you in that article is that only the facial area is computer generated. They filmed an actress, scanned her facial expressions, and plonked it straight back on her face.
http://technology.timesonline.co.uk/tol/news/tech_and_web/article455793…
Higher-res video here:
http://www.awntv.com/videos/image-metrics-emily-project
They claim it's a leap over the uncanny valley problem, but it's a sort of a cop-out solution I think, since they're basically scanning a face and rendering it again.
It reminds me, Team Bondi has some facial scanning technology that they've patented (revealed in a job news item). Would be curious to see that when L.A Noire sees its release.
God dammit, why are Pixar
God dammit, why are Pixar the only people who seem to be able to do eyes properly?
Her face smiles, but her eyes don't. It's missing all those slight varied tiny movements our eyes make that we don't notice. You do however notice those movements once they're gone.
The material and the render is actually epic though, just the eyes and the lips really ruin the animation for me.
Dane Brennand - Texture Artist
Yes ... but !
They are saying that this is what the technology is supposed to capture; a more 'continual' and subtle control over the movements and expressions - that is eyelids, lips, etc. I think as far as digital processes go it's a leap forward. It's not very realistic, (try watching it with the sound turned off!) however the speed and cost with which they can produce the quality shown is pretty awesome.
Digital solutions become cheaper and cheaper as time goes on, so this method of image capture and processing is going to be the status quo soon, even if it might be considered non-comparable to some other current methods that require more human/analogue control at the moment. What we *should* compare it to is the old methods of capture and purely digital-recreation ... which are pretty crap.
Well, it's not as 'non-trivial' as you might expect.
Detecting common features from a 'transient' image (like video) and then using those features as 'markers' from one frame to a next, then mapping both the geometric data and then the texture data onto the geometry (rather than motion capture, which has physical markers present in the image data to do all this) is a pretty cool little application of transform methods and signal processing. What benefits this has would depend on the system of capture used, how much data you can collect and process etc etc
I'm certain it has been covered in research already. However, obviously there are many ways to implement it an 'image metrics' is simply one system devised for this purpose. I really would have to take a closer look at it to work out whether it gets my tick of approval or not.