Meanwhile elsewhere
http://www.youtube.com/watch?feature=player_embedded&v=fAsg_xNzhcQ
(will be published in an online book, gpu-gems style in 2-3 months, its editor is a colleague of mine, so I can testify it's real and realtime :) )
Posted on 2011-08-02 19:01:44 by Ultrano
That's an interesting use of voxel technology.
I've seen a similar approach to indirect lighting in CryEngine 3 a few years ago, but I don't know how they do it exactly. It is volumetric, but not sure if they store it with a voxel-like approach:
http://www.youtube.com/watch?v=FqeXuO2AlEE
http://www.youtube.com/watch?v=_rkLEsR9bRs

http://www.incrysis.com/index.php?option=com_content&task=view&id=812&Itemid=1

Edit: Upon closer inspection of the paper and presentation, it appears that they encode the volumetric data using spherical harmonics.
Posted on 2011-08-03 05:06:50 by Scali
Scali, I wanted to revisit this one.
You stated that NURBS are not really non-polygonal data.
But this is in the context of a pixel shader! We're not digitizing the curve at whatever tesselation factor, we can SKIP the vertex shader stage completely.

Anyway, it turns out that the technique being applied is not based on that stuff.
It's merely a pixelshader based voxel engine, and it's a lie - look at the instancing, look at how the instances all have the same orientation, etc.

It got people talking, it made someone some grant money for talking gibberish and sounding smart.
Posted on 2011-08-18 04:33:59 by Homer

Scali, I wanted to revisit this one.
You stated that NURBS are not really non-polygonal data.


Indeed, I meant that NURBS are (usually rectangular) patches. So your control mesh is basically still a polygon mesh.
But I was talking about Pixar, not Euclideon. These guys use voxels with sparse octrees. Common trick, really.


But this is in the context of a pixel shader! We're not digitizing the curve at whatever tesselation factor, we can SKIP the vertex shader stage completely.


Depends on how you look at it... I think Pixar rather sees it the other way around: you skip the rasterization stage, so whatever you do is more equivalent to a vertex shader in D3D/OpenGL context than to a pixelshader.


Anyway, it turns out that the technique being applied is not based on that stuff.
It's merely a pixelshader based voxel engine, and it's a lie - look at the instancing, look at how the instances all have the same orientation, etc.


Yes, that's what I said as well. If they want to show that it is a viable game technology, they need to make a game demo with it.
One that includes proper shading, shadowing, post-processing, animation, and more realistic scenes.
This mainly shows off that tracing through a sparse octree in realtime is possible, and you can instance that octree as much as you like... But others have shown this before these guys. And some have even included basic animation (which also indicates that it is probably not ready for games just yet... a single isolated character can barely render at 36 fps... so I doubt that a bunch of such characters in an actual scene would still get playable framerates: http://www.youtube.com/watch?v=tkn6ubbp1SE).
Posted on 2011-08-18 05:49:16 by Scali