quote:
Yann: http://www.donw.co.uk/home/downloads/SHTest.zip - SH rotation code using Ivanic/Ruedenberg algorithm. Full references provided.
Thanks AP, much appreciated. I''ll take a closer look at those rotations tonight.
quote:
Yann can you post those links you mentioned earlier, please?
Yeah, I''ve been very busy the last couple of days. OK, before going anywhere into more advanced solutions, you should make sure you perfectly understand the mathematical principles behind radiosity: the basic GI equation, the simplifications used for the diffuse interreflection model, differential angles and patches, form factors, etc.
That Hugo Elias thing is nice as a quick start, but if you want to dive deeper, you should try to get an accurate ''real'' solver running first. You can then extend it to accomodate various other types of algorithms, always comparing their visual results and performance to your reference implementation. The implementation can be a full matrix solver, or a very basic progressive shooter. I''d suggest trying it on a standarized scene, such as the
Cornell box. You can easily use the precise data sets to verify the proper functioning and quality of your system, as well as compare it with other solvers.
Once you have the thing running, the next step would surely be an iterative solution, in order to improve performance and reduce the memory requirements. I''d suggest trying progressive refinement (PR) radiosity next. Several different approaches for PR radiosity exist. You should try a maximum energy shooting approach, with implicit ambient term. That''s pretty much the industry standard: very good quality, and more or less fast. The form factors can either be determined by raytracing (best), hemicubes (fast) or other (hybrid) approaches. Test a few different methods, and compare the results. This will give you a ''feeling'' for the various algorithms.
From here on, you can start playing with probabilistic solutions, such as (Quasi-) Monte Carlo radiosity. Again, lots of different variants exist. A very advanced form is wavelet radiosity. Interesting in theory, although not that useful in reality. You should be very confident with your math before attempting this one.
Practically all advanced radiosity algorithms are presented in the form of (math-heavy) academic research papers. Here is a short list, just Google for the exact title match. Some might require an ACM account, I''m not entirely sure.
Radiosity introduction:
"Introduction to Global Illumination" by Paul Heckbert
"Analysis of Radiosity Techniques in Computer Graphics" by Bernard Kwok
"An Empirical Comparison of Radiosity Algorithms" by Paul Heckbert et al.
Iterative and Monte Carlo solutions:
"Iterative Methods for fast Radiosity Solutions" by Shirley, Baranoski, Bramley
"Hierarchical Monte Carlo Radiosity For General Environments" by Pérez et al.
"Hierarchical Monte Carlo Radiosity" by Bekaert, Neumann et al.
Form factors by hemicubes:
"Anti-Aliased Hemicubes for Radiosity" by Chandran, Kiran, Phani
"Fast Texture Based Form Factor Calculations for Radiosity using Graphics Hardware" by Nielsen, Christensen
Wavelets:
"Wavelet Radiosity" by Gortler, Schröder, Cohen
This should get you started. Also have a look at this forum''s FAQ, I have collected a few related links there.
While you''re at it, don''t forget to take a deep look into high dynamic range (HDR) imaging. Radiosity and HDR are tightly interleaved, and a good HDR remapper (exposure) is absolutely vital to get a pleasing visual result from a radiosity solution.
Last but not least, if you need some example code, take a look at
Fluid studio''s radiosity processor. A nice basic PR solver, comes with full source.
