Bih - Bsp2

[Morbid Angel]

Hey everyone. Have an interesting read, and Vlado, any thoughts?

they claim it to be faster then regular BSP by a factor of 2. Now I know vray does not use typical BSP, but how does it compare to BIH?

http://en.wikipedia.org/wiki/Boundin...rval_hierarchy

http://graphics.uni-ulm.de/BIH.pdf

[Paul Oblomov]

BTW - I found hat there's so much new realtime raytracers outside... Maybe they based on that technique

[Morbid Angel]

well this bih can handle an extremely large amount of triangles, which I dont see the point to use in games.
Anyways I hope vlado can clarify

[vlado]

The advantage of the bounding interval hierarchy is that it is very fast to update the strcture dynamically if you have moving objects and need to rebuild it often (e.g. for each frame in a game).

However, due to the fact that it does not split the triangles, it performs poorly for scenes with large triangles across large spaces (e.g. walls with only a few triangles) and may fail to subdivide the geometry in an optimal way. In a particular implementation that we looked at, ray casting though the BIH was as much as 5 times slower than a regular kd-tree.

In a real-time sitation though, the savings you get from the linear build time for BIH compared to a kd-tree may be enough to offset the slower tree traversal.

For offline raytracing, where the build time for the tree is negligible compared to the actual rendering, BIH does not make much sense, except perhaps as the initial coarse subdivision in a hybrid BIH/kd-tree solution.

Best regards,
Vlado

[Morbid Angel]

so you are saying there is no real point to it? well why implement it into mr then?

[vlado]

As far as I know, mental ray bsp2 does not use pure bounding interval hierarchy; further on there are cases where bsp2 is slower than the regular bsp...

Best regards,
Vlado

[Morbid Angel]

well I just hope they didnt swap the bsp for bsp2 completely. And left it as an option.