• Nvidia crosses 'Uncanny Valley': Digital Ira loves the Titan
    44 replies, posted
[media]http://www.youtube.com/watch?v=5d1ZOYU4gpo[/media] The fun begins at 8:40 Basically nVidia found a way to condense 32 GB of captured facial animation data into around 400 MB of mesh data, animating it takes 2 TFlops, around half of what the Gforce Titan provides. They say they're out of uncanny valley but if you ask me, while looking incredible, it still looks kinda uncanny, especially becaus the pink skin inside your eyes, around your eyeballs is missing.
8minutes 40seconds. Holy. Shit.
The mouth movements are way off for me
what I love about increased realism is that more style can follow. Imagine a game that looks like it's in 2D but is rendered in 3D, and I don't mean cel shaded style, I mean 3D that looks like actual 2D, complete with small jitter in lines, more human looking shading and stuff like that. the Talking looks creepy tho.
[QUOTE=Whomobile;39988973]what I love about increased realism is that more style can follow. Imagine a game that looks like it's in 2D but is rendered in 3D, and I don't mean cel shaded style, I mean 3D that looks like actual 2D, complete with small jitter in lines, more human looking shading and stuff like that. the Talking looks creepy tho.[/QUOTE] This, GCI that resembles traditional animation is something I've always wanted to happen. Paperman was a step in the right direction and looked amazing, though you could clearly see it was made in 3D sometimes.
Nope, still looks creepy as hell.
Hang on, the robots from I Robot didn't fall into the uncanny valley? [I]Suuuure.[/I]
That goddamn smile..
All i wonder now is how long it took to create that face and animations.
400 MB per character is still not very feasible for games and all of that stuff but I guess it's a really nice tool. The downsize is that it makes the animation dependend on that capturing technology.
I can't wait until my triple a games take up 600 gigabytes of space.
i'd like to see this rendered at 24fps. i bet part of the weirdness of the detail and fluid animation is because we're viewing it at 60fps (or 60fps slowed down to whatever it is youtube runs at), in the same way that most people think the hobbit looks weird
[QUOTE=ChestyMcGee;39989542]i'd like to see this rendered at 24fps. i bet part of the weirdness of the detail and fluid animation is because we're viewing it at 60fps (or 60fps slowed down to whatever it is youtube runs at)[/QUOTE] (30fps)
Take that PS4!
[QUOTE=Killuah;39989327]400 MB per character is still not very feasible for games and all of that stuff but I guess it's a really nice tool. The downsize is that it makes the animation dependend on that capturing technology.[/QUOTE] Maybe not now, but think of it this way. The absolute highest end video card in 2005 was a 7800GTX with 256MB of VRAM. Now we have a Titan with 24x as much memory. Eventually 400mb per character won't be anything.
Half-Life 2 crossed the Valley, at least to my crappy eyes.
[quote= guy]what would it be like if we could use this for telepresence, so instead of videoconferencing, our words are translated into instructions which you would express or animate the avatar on the other end.[/quote] can we get a drug test for this guy?
[QUOTE=darth-veger;39989306]All i wonder now is how long it took to create that face and animations.[/QUOTE] Well the animations process wasn't too different from the one used in L. A. Noire so it can be efficient I'm sure.
[QUOTE=Brt5470;39989648]The absolute highest end video card in 2005 was a 7800GTX with 256MB of VRAM. [/QUOTE] No, there was a faster 512MB version released later that year.
[QUOTE=Silikone;39989795]No, there was a faster 512MB version released later that year.[/QUOTE] Just looked it up again, I guess there was indeed a 7800 GTX 512Mb version in november of that year.
[QUOTE=Durrsly;39989686]Half-Life 2 crossed the Valley, at least to my crappy eyes.[/QUOTE] Half-Life didn't cross the valley, it was still far too obvious that the characters are fake to be creepy.
The ears are... off?
[QUOTE=Hiccuper;39991687]Half-Life didn't cross the valley, it was still far too obvious that the characters are fake to be creepy.[/QUOTE] L.A. Noire is somewhere in the valley. [video=youtube;q2EG5J05048]http://www.youtube.com/watch?v=q2EG5J05048&list=PL1AF1F0CC435D69B7&index=11[/video]
[QUOTE=Zoo;39992357]L.A. Noire is somewhere in the valley.[/QUOTE] the problem with L.A. Noire is that while the facial animations were great, the body movements didn't match. It was like the face was an entirely different performance than the physical movements of the character's body.
They still cant get him to smile convincingly
Also, I think the lack of red tissue around the eyes was probably because since the eyesockets tended to be shadowed, so you wouldn't be able to see it anyways.
It's the eyes. It's always the eyes.
[QUOTE=Zoo;39992357]L.A. Noire is somewhere in the valley. [video=youtube;q2EG5J05048]http://www.youtube.com/watch?v=q2EG5J05048&list=PL1AF1F0CC435D69B7&index=11[/video][/QUOTE] L.A. Noire is in the valley? When I played it didn't give me that impression. The facial expressions and features (to me) look real as hell, but what it lacks is mesh quality more often than not because of the way the mesh is created (space between the teeth is flat since the cameras can't tell the mouth is hollow). It never came across to me as creepy though.
If you're looking for good facecap look at Halo 4.
I do think LA Noire is very much inside the uncanny valley. As with any kind of [b]tracing[/b], there's only so close to reality you can get with mocapping, plus you have a considerable problem with making all that animation cycle and interact with the environment properly. Cinematography can use specific angles and technqiues to make it less obvious the actor's playing to a greenscreen, but if you give the player all that camera direction, you can't count on that ground work. Most blatant example of that effect is maintaining [b]eye contact[/b] with the player. Even after a decade HL2 is one of the few FPS that taught their NPC to look at you and face you even when your character's jumping around. Personally I think we'd be better off with more [b]stylization[/b], not just because it's way more cost-efficient and we're already reaching something of a money- and hardware-hungry "saturation curve" in our attempt to leave the uncanny valley (most people judge graphics by how big the rock textures are anyway), but also because it means we'll have to rely on custom animation rather than uncanny mocapping that'll never be perfect anyway. It's already putting limits on the industry like voice actors being excluded from audition because they can't mocap crazy acrobatics and apparently devs like to use them for both. I find it incredibly bizarre that we're living in an environment where Crytek boasts their stuff's so awesome it melts PCs. Isn't a good, optimized engine supposed to achieve the opposite effect? Another problem is that the closer we're getting to graphical realism, the more an absence of realism in all the other game aspects will stick out and we'll dive into [b]yet another uncanny valley[/b] where it's again difficult to escape from.
Sorry, you need to Log In to post a reply to this thread.