Voxel vs BVH back

(L) [2012/02/15] [ost by spectral] [Voxel vs BVH] Wayback!

Hi All,

To be honest I have no experience with the Voxel rendering techniques (except some theory) and so I would like to know your advice.
My current engine use a BVH but why not using Voxel today ?

Is there some advantages in one method compared to the other one ?

(L) [2012/02/16] [ost by beason] [Voxel vs BVH] Wayback!

because it's a lot easier to represent a polygon using polygons instead of voxels?

(L) [2012/02/16] [ost by spectral] [Voxel vs BVH] Wayback!

It seems more natural for us.

But in the reality objects are not made of polygons but of molecules by examples... so Voxel looks more realist to me [SMILEY :-)]

But for a computer, what looks more easy to process [SMILEY :-P]

(L) [2012/02/16] [ost by toxie] [Voxel vs BVH] Wayback!

Voxels should usually be a bit more speedy to process than f.e. a BVH and tris.. But then on the other hand, as beason mentions, triangles are common, so pre-voxelizing these and using them can be less speedy than using them directly.. Also the thing with voxels is, that as soon as you have large/huge scenes (and thus lots(!) of empty space/voxels), the pain and tweakery starts and the voxels are not just a simple grid anymore (depending on your definition of voxels, but i guess thats what you mean), whereas the BVH can handle this stuff out-of-the-box..

(L) [2012/02/16] [ost by spectral] [Voxel vs BVH] Wayback!

That's right... I have also think to the large scene problem with non uniform atmospheric effects... impossible to handle with voxels [SMILEY :-P]

I was more thinking to speed some interesection test with it, not all [SMILEY :-P]

(L) [2012/02/16] [ost by toxie] [Voxel vs BVH] Wayback!

if its rather compact (not that many empty voxels or even overall voxels), then performance will be awesome, as there is not much data per traversal step, and the traversal code itself is simplistic, too..

(L) [2012/02/16] [ost by graphicsMan] [Voxel vs BVH] Wayback!

It seems though that the (virtual) grid will need to be extremely fine to capture smooth surfaces unless the cells describe isosurfaces via corner values and gradients; at that point it becomes less cheap too [SMILEY ;)]  For any real scene you need an adaptive grid, octree, BVH, or kdtree+leaf box to make this practical.  I do wonder, however, if beam tracing or wavefront tracing might become more practical with a voxel setup.  Anybody looked into that?

(L) [2012/02/16] [ost by ingenious] [Voxel vs BVH] Wayback!

I guess it also depends on what scenes you want to render. Architecture, cars, most objects in product design, these can all be represented well with triangles and/or higher order surfaces. My personal problem with voxels is, as graphicsMan mentioned, smooth and flat surfaces - something that you can represent well with a few triangles can cost kilobytes of voxel data. I personally haven't seen a voxel demo that shows typical interiors/exteriors. For plants and forests, though, voxels should rule, especially with the intrinsic level of detail.

(L) [2012/02/17] [ost by paulusul] [Voxel vs BVH] Wayback!

Not to share the kool-aid, but I think that unlimited detail shows that voxels have not yet got the amount of attention they deserve.

Of course UDs claims are hilariously unbelievable to any one with any knowledge of computers, but at least they have nailed smooth surfaces.

for any one that does not know UD, a recap.
[LINK http://www.youtube.com/watch?v=UO5tFZC8wWU]

Sorry if that was way offtopic

(L) [2012/02/20] [ost by toxie] [Voxel vs BVH] Wayback!

and we are still waiting for some real life proof of what they're doing, and not just videos.. :/