(L) [2009/10/07] [sayinghai] [ Smoother Volume Data Sampling??] Wayback!Hello Ompf,
Long time reader, first time poster. Just like to start out by saying that on the whole OMPF is pretty awesome. I've come across a few names that I remember from flipcode.
Anywa, I'm working on a custom volumetric rendering solution using PRMan. I have a DSO that loads up some volume data an samples it as shader marches rays through it. To get rid of the grid, I added a random number between that fits into the size of a data cell. It's just a uniform random number. It got rid of a the grid, but now it looks a bit noisy. It's not terrible....however, I was wondering if someone had some ideas on how to reduce? Or maybe a better sampling strategy? The data is being summed across a 3x3x3 block using a Gaussian function. Unfortunately, I'm under NDA and blah blah blah some big studio...lawyers, etc.. - so I can't post any images.
Any thoughts/ideas would be appreciated.
Thanks.
(L) [2009/10/07] [toxie] [ Smoother Volume Data Sampling??] Wayback!just always step in uniform ray steps, so basically 1/n.. as this leads to banding when looked at the "full picture" and not just at a single pixel/sample level, you should provide an initial offset for each ray in the range of [0..1/n).. this can be a random number or even better the third/fifth dimension of your screen/lens sampling (QMC?!)..
(L) [2009/10/07] [calamari] [ Smoother Volume Data Sampling??] Wayback!Here's the corresponding paper, see Fig. 2 c):
[LINK http://www.uni-kl.de/AG-Heinrich/MediaMLT.pdf]
(L) [2009/10/07] [spacerat] [ Smoother Volume Data Sampling??] Wayback!here another one, maybe a bit more general:
[LINK http://viscg.uni-muenster.de/publications/2009/RHRL09/?clang=1]