GTA IV team creates Virtual Human

I wouldn't call that "GTA IV" team's work. Just because they provided mo-cap data for GTA4 doesn't mean they are GTA team. It may be rendered real-time but the amount of data is probably impractical for games. You can stream hundreds of megs of data per second but where are you going to store them? Even if you compress them 100:1, 1GB of disk storage gives you 15minutes of data. Pretty tight for a game (and only practical for cutscenes).
 
We don't know if it's the same technique AMD uses (they are just using AMD hardware) and voxels are no magical way to squeeze data. http://www.vfxworld.com/?atype=articles&id=3723&page=1 article doesn't mention voxels at all.
No I think they are just using polys and they have built shaders to recreate the lighting functions of her skin from the data that they have collected.

But the demo is mostly about animation. They learnt new subtleties in the behaviour of skin and constructed a very detailed animation rig.

Voxelisation is an independent technique for capturing both the geometry and the lighting properties of a face.

So as I understand it, the eventual aim is to generate animation rigs from real people and separately capture (or generate) geometry. Voxelisation is heavily compressible and extremely suitable for LOD - but as far as I can tell is difficult to animate. So the remaining problem is to get from an animation frame to some kind of transformation of the voxel data, before rendering it.

Voxelised data appears to be better suited to being rendered by ray-casting/tracing, from what I've read - so the other major issue is that game engines are simply not geared towards this technique.

id Tech 6 is supposedly going to limit voxelisation to the landscape/structures and the characters/moving objects will remain as polygons.

So it's still a way off. Considering that a lot of people are still scratching their heads over the fine detail of MegaTexture, I suppose it's fair to say it's gonna take some geniuses to blaze a trail to fully homogenise this kind of rendering of people.

Certainly Jules Urbach is attacking voxelised+animated people. Huge compression ratios are a key part of that. The way he's come out of nowhere does give me some confidence :D

Jawed
 
Back
Top