Occlusion culling shadow volumes

Started by
5 comments, last by krausest 18 years, 7 months ago
I am wondering about the CC Shadow Volumes code here: http://gamma.cs.unc.edu/ccsv/ Is it viable to do real time per-poly occlusion culling in this case? If so would you do this by testing whether an object is inside another object's shadow volume? Would you order the shadow volume creation by doing a depth sort from the camera position to optimise this? Does anyone know of any games using this technology?
Advertisement
Bump. I really need some answers on this one, soon enough I'll be implementing this stuff for real, I'd rather understand it properly.
I havn't read the paper yet (but thanks for the link!), but I would never sort by polygons.
I would, (if using shadowvolumes) use per-"object" occlusion-culling, if I knew that the scene needed it.
The images on that page are kindof extreme, and usually the overdraw isn't that bad.

Good luck!
I asked the same question some time ago
( http://www.gamedev.net/community/forums/topic.asp?topic_id=330245 ), but didn't get the definite answer I was looking for.
It sounds as if it would pair nicely with the other technique mentioned in that thread, because both need a shadow map to optimize stencil shadows.
i wouldn't use this technique, it looks like it eats fillrate and cpu time like candy

Altough it's an interesting idea, it's not very scalable
Stencil shadows are already becoming outdated.. even when mixed with other techniques
I remember reading that paper a long time ago. I think the impression I got was that the increase in batch count would hurt performance considerably. Not sure though, might have been something else. I believe they showed that the cpu overhead was less expsensive than the fillrate saved, but then again you are commonly cpu bound now days, and fillrate wasn't so plentiful back then.

Personally I wouldn't say that shadow volumes are outdated. They have been around a long time, and they do have problems, but I believe they are all solvable. The major ones people seem to list are using non-2-manifold geometry, the cpu overhead, and of course fillrate. the first is solved, the second I feel should be solvable when paired with a gpu like the Xenos in the xbox360 (ability to write to memory) and also with the 360, having the z/stencil ops done on the DRAM itself should solve the fillrate problem.
So I wouldn't say they are outdated.
Quote:Original post by LogicalError
Stencil shadows are already becoming outdated.. even when mixed with other techniques


So what is your solution?
I've tried PSM, TSM and had to find out that a single shadow map doesn't deliver constantly good quality.

I've build a multiple shadow maps demo that uses PSM for each shadow map, but performance isn't that great (at least for my demo it's slower than stencil shadows) and quality is quite good, but noticeably worse than with stencil shadows.


This topic is closed to new replies.

Advertisement