I recall an article that was discussing a BVH splitting algorithm that whent along the following lines:
1  Pick any two primitive
2  Compute two centroids from those two triangles
3  Gather two lists of all the primitives that are nearest those two centroids
4  If the list members changed since the last iteratiop then
4b  Compute two centroids from the members of the two lists
4c  Goto step 3
5  Otherwise the BVH is splitted at this level. Continue for the next levels.
Does anybody know an article that discusses such BVH splitting method?
Looking for Nearest Neighbor BVH splitting algo

 Posts: 12
 Joined: Tue Dec 06, 2011 8:18 am
Re: Looking for Nearest Neighbor BVH splitting algo
(Shadow007 here, just changing my nickname ! Pleased to find a now functional again ompf forum!)
It seems it's not the paper you want, but your algorithm explanation reminded me of the "Fast Agglomerative Clustering" one by Walter; Bala et al :
http://www.cs.cornell.edu/~kb/publications/IRT08.pdf
It seems it's not the paper you want, but your algorithm explanation reminded me of the "Fast Agglomerative Clustering" one by Walter; Bala et al :
http://www.cs.cornell.edu/~kb/publications/IRT08.pdf
Re: Looking for Nearest Neighbor BVH splitting algo
Thanks. You're right. Not the paper I'm looking for but interesting nevertheless.
Re: Looking for Nearest Neighbor BVH splitting algo
This method seems to be an application of a more general principle called "vector quantization", which I encountered in the context of data compression. These specific buzzwords might help narrow down your search. (And in case you already knew this trivia, then please don't feel offended that I pointed out the obvious.)