Most of that research paper is above my understanding. However, I think I can
make a *few* safe comments, regarding POV-Ray.

The paper speaks of 'photon mapping' as an all-inclusive rendering technique
there. POV-ray also has 'photon mapping'--but the *meaning* is somewhat
different. POV-ray's use of 'photons' is solely to produce caustics (from light
shining through transparent glass, etc), while 'global illumination' is done
with the 'radiosity' feature. (Both photon-maps and radiosity 'maps' *can* be
pre-generated and saved, then used for the final render, if desired.) But
POV-ray doesn't use so-called photon-mapping for its *entire* rendering
pipeline. I think the paper is describing a different approach.

The overall technique described on the FIRST page of the paper has great
similarity to POV-ray's radiosity feature-- you could almost replace 'photon
mapping' there with 'radiosity', and it would make sense.

In the past, I *have* seen newsgroup posts here about various 'forks' or
unofficial versions of POV-Ray that might use some kind of photon-mapping or
'Monte Carlo' approach to overall rendering; you could try doing a search.

Over the years, it has been said many times in the newsgroups that POV-ray's
overall raytracing technique cannot make use of a GPU's power, just the CPU (or
multiple CPUs, as it does work with parallel processing.) I don't know the
technical reasons for non-GPU use, but I think raytracing itself is difficult to
implement. Of course, that may have changed; I've seen other research papers
that talk about 'real-time raytracing.' It has even been discussed here in the
newsgroups, from time to time.

Sorry to say that I don't know much at all about k-d trees, so I can't help
there.

Post a reply to this message