PDA

View Full Version : Light Space Perspective Shadow Maps



knackered
11-17-2004, 12:53 AM
Anyone had any experience with them?
http://www.cg.tuwien.ac.at/research/vr/lispsm/
Haven't read the pdf yet, but the demo exhibits some 'swimming' artifacts, which I suppose would be reduced as you increase the shadow map resolution.

davepermen
11-17-2004, 04:42 AM
looks quite nice afaik. still downloading the stuff to watch more than the pics, though :D

zed
11-17-2004, 09:53 AM
Haven't read the pdf yet, but the demo exhibits some 'swimming' artifacts, which I suppose would be reduced as you increase the shadow map resolution
not really gonna solve it, but from when the light view changes even minutely u get different pixels in the SM, i havent implemented Lispsm but with standard sm, u can solve this by rounding (the shadowmap position? sorry about my english)

A027298
11-17-2004, 05:58 PM
I made quite good experiences with Trapezoidal Shadow Maps (http://www.comp.nus.edu.sg/~tants/tsm.html)

stefan
11-17-2004, 10:45 PM
Regarding TSM: For certain positions of the light, the 2d view of the trapezoid becomes almost a rectangle. How do you handle those cases? I didn't find anything in the paper for that case. If I try to create a trapezoid for such a configuration the wasteage is of course very big and quality is even below standard shadow maps.

About LISPSM: LISPSM depend on the convex hull of the view frustum and the light position. Wasn't that of the things that caused severe problems for the original PSM approach?

stefan
11-17-2004, 11:12 PM
Sorry I was a bit unclear. The problem is not the computation of the convex hull, but that the convex hull must be visible from the light view.

A027298
11-18-2004, 12:11 AM
Regarding TSM: For certain positions of the light, the 2d view of the trapezoid becomes almost a rectangle. How do you handle those cases? I didn't find anything in the paper for that case. If I try to create a trapezoid for such a configuration the wasteage is of course very big and quality is even below standard shadow maps.
Yes, that happens for the dueling frusta case. However, all the shadow map techniques (PSMs, LISPSM, ...) have that limitation. For the TSM case I wouldn't say the shadow quality goes below standard shadow maps. Taking a look at http://legion.gibbering.net/projectx/paper/shadow%20mapping/#_Trapezoidal_Shadow_Mapping_Overview where this problem is addressed to some extent might be interesting.

About LISPSM: LISPSM depend on the convex hull of the view frustum and the light position. Wasn't that of the things that caused severe problems for the original PSM approach?
Yes, both depend on the convex hull of the view frustum, light position and scene bounding box which might introduce further discontinuities in a dynamic scene. TSM only depends on the intersection between the view and the light frustum which are scene independent.

stefan
11-18-2004, 01:15 AM
Thanks for your link, I'll have to take some time to work through it.
At the first sight, my problem seems to be that the point used for the projection (and used to define the side lines of the trapezoid) doesn't move away from the trapezoid (it should go to infinity for the duelling frusta case, then I'd get almost a rectangle and the wasteage would be less). As the point stays close to the near front line there's a lot of wasteage at the distant edge of the trapezoid.

I've implemented PSM as described in the GPU Gems. I don't see problems for the dueling frusta case. The quality is also in every case better than TSM and LISPM (which might be a fault of my scene and/or my implementation). But I must admit that I've spent much time on optimizing PSM (unit cube clipping and adjusting z-Near). Also PSM in that version only depend on the view frustum, but not on the convex hull of the light and the view frustum (Though it's a trick that might go wrong, this is why nvidia takes only a fraction of the z near distance to the unit cube in their demo...)

Dirk
11-23-2004, 08:49 AM
Originally posted by A027298:
Taking a look at http://legion.gibbering.net/projectx/paper/shadow%20mapping/#_Trapezoidal_Shadow_Mapping_Overview where this problem is addressed to some extent might be interesting.
Interesting read, especially the last sentence:
the trapezoidal algorithm is patent pendingwhich is really bad news. :( Where did you get that from, was that a part of the paper that I missed? If that's true, the whole thing is useless.

Thanks

Dirk

A027298
11-23-2004, 05:05 PM
the trapezoidal algorithm is patent pendingThere is a note about that on the TSM project page (http://www.comp.nus.edu.sg/~tants/tsm.html).

MickeyMouse
11-23-2004, 11:00 PM
Have you seen Irregular Shadow Mapping (http://www.cs.utexas.edu/users/billmark/papers/ism-TR/) ?

Is summary - quality of Shadow Volumes at the cost of Shadow Mapping (with some const multiplier).

That's shadowing method we're waiting for, but we must convince nVidia and ATI to implement Irregular Depth Textures on their hardware first ;)

Sorry for OT :D