>
> > You're right that it is more specifically defined as a KD-Tree.  It
> > would be a fair amount of work at this point to fix all the references
> > so I've left it as BSP-tree.  It seems that BSP trees also have a
> > third definition and that is that the splitting plane is axis-aligned
> > AND it is put in the center of the cell it is splitting.  If there is
> > enough interest in the patch I could correct the naming.
>
> If I were you I would change the naming just to make sure to avoid
> confusion when looking for algorithms and insights in the literature.

Well, its not always referred to as a KD tree in the literature, for
example:
http://www.acm.org/jgt/papers/HavranKopalBittnerZara97/

The algorithms in Mental Ray do indeed use a KD tree (at least according to
a reference in the paper: http://www.cgg.cvut.cz/~havran/phdthesis.html) but
they refer to it as a BSP tree nonetheless.  I remember seeing other
references to raytracers using orthogonal or axis-aligned BSP trees but
don't have them on hand.

> I noticed that you have also a "competitor" ;-), and that you've already
> discussed a bit on your respective implementations, any thought of
> "merging your results taking the best from both"?
>

That would be an interesting idea, though I'm not sure how practical.  Since
we both seemed to implement from the same paper there is much overlap in the
implementations.  IMHO, my current implementation at least has a more
modular and flexible interface.  His patch has a more efficient construction
phase but allocates larger arrays.  My previous testing also seemed to show
that my patch has better rendering time performance, though we really need
an impartial judge to do the comparison and analysis :)

Post a reply to this message