Valid XHTML 1.1! Valid CSS!

Photo of Andy Scukanec

Research

BSSRDF Implementation in RTGI

Currently, I'm working with Steve Marschner to try and implement his BSSRDF paper into the RTGI framework that is heavily used by the PCG. The framework has some nice properties like being able to distribute the rendering task over the cluster, a large base of correctly implemented materials, geometry, acceleration structures, etc. Unfortunately, it was not built with the idea that people will want to use non-local information during the rendering task. As a result, we have had to design a small space in which we could develop our implementation. This wasn't easy though, since it needed to:

What resulted was the PAX rendering package which was a non-necessary subset of RTGI. PAX, RTGI, and participating media deserve a lot more thought on how to do them right, since PAX is admittedly just a quick fix.

July 20, 2004: Thus far, I have managed to get the single scattering term implemented correctly. I am off by a simple scale factor on my multiple scattering term, (on the order of 1.3 to 1.5, but its constant). I should be able to hunt down the error in short order. Of course, when I say these things are correctly implemented, what I really mean is that my Matlab and Java code agree for a particular scene or two. Its completely possible that both are pooched, but I'm hoping that is not the case. I'll be adding more and more difficult test scenes later (ie - off normal results, changing sigma_s and sigma_t, etc).

July 18, 2004: I have just verified my Java implementation against my Matlab implementation. They match up perfectly, but now I need to have the same scene rendered by the full Monte Carlo renderer and make sure that I can match up with external results. I also need to try out more test scenes and verify my results with Hongsong's MC renderer. I also want to make some modifications which will allow easier use of the PAX architecture. Right now the only way to associate a PAX shader with a material is by hardcoding such an association into the code. This information should be placed in the input file, as well as any parameters to that shader for that material.

Notes on implementation [pdf]

Sandbox page [CUCS/PCG Only]