Hi, after trying out several Shadow mapping algorithms, I settled for Paraboloid/Dual paraboloid shadow mapping. However, I have a problem - all the shadows I get are a bit "curved". This is not much of an issue for figures etc., but it is pretty noticeable for vertical columns casting shadows on the walls. I'd like to ask: Are the curved shadows a normal feature of Paraboloid shadow mapping? Or is it just a bug in my code? Thanks.

It depends. Are you just implementing Brabec's original algorithm (with all of the math being done in the vertex shader?) If so, then you're going to get curved shadows in almost every case. If you haven't already, I'd suggest you look at our paper:

http://portal.acm.org/citation.cfm?id=1183331

I need to find out if I'm allowed to distribute that directly, rather than forcing people to have ACM memberships. Let me know if you can't get a copy. Anyways...

People will point out that you need to have your geometry tesselated to avoid that problem, but that's only really true for your shadow casters. You can keep your receivers (eg, walls) low-poly (and it sounds like that's the case you have), by moving the lookup math into the pixel shader. There are some other things to think about, but in our experience DPSM is a completely viable solution for omnidirectional shadow mapping (given a few constraints).

Thanks for answering. Yes, I do the math in VS. I sent you a PM.