I'm trying to implement shadow mapping on top of the well known Catalin Zima deffered rendering article. When you get to the point where you rendered the shadow map, and are now trying to check whether the distance from the pixel you're looking at is lower than the distance written in the shadow map, what is the formula for getting the correct texture coordinates in the shadow map.

In one sample this is used:



The shadow mapping sample uses this:


And a render monkey sample uses this:


The shadow mapping sample and the first sample use an orthographic camera as the light camera, while the render monkey uses a normal fov camera (i think).
How does the type of camera projection affect these shadow coordinates, and what's the math behind this?

So, it really comes down to projective texture mapping. You can read a pretty good article on it here.

In my implementaion I've based on implementation of deferred shadows, that use an ortho camera. I don't know the really math behind this, but for reference you can take a look at a sample i've made, check it on my signature link. The sample technique, is the most simplest way I've found, I've made some improvements to final version that i've used in my game. But both have some problems, but for now I won't take it again. Sorry about english

Yeah, thanks I already looked at your sample. I found it a bit confusing, especially the shader code and the calculate frustum code, but I experimented a bit with it. I think I'll switch my code to use an ortho camera also, I think I'm getting bad results due to shadow map inaccuracies. By my calculations the shadow map I got from the light fov camera is using about 1 / 5 of the whole range.

I'm currently doing this:

RenderShadowMap();
DrawSceneUsingDefferedRendering()
DrawLights();
CombineMaps();

The draw scene renders color, normals, and the occlusion into g-buffers, and the the CombineMaps() combines them all.
Is that ok?

Does anyone have a clue what could be causing this:

http://a.imagehost.org/0364/Abandoned_2009-08-27_17-58-56-62.jpg
http://a.imagehost.org/0573/Abandoned_2009-08-27_17-59-00-78.jpg

When I move a little closer or farther it's normal:

http://a.imagehost.org/0179/Abandoned_2009-08-27_17-59-06-89.jpg

Also this:

http://a.imagehost.org/0373/Abandoned_2009-08-27_17-59-52-17.jpg

The spikes are not in the terrain!
I can post my code if it would help.

That banding *might* be caused by 2 things:

I'll try splitting my floats, thanks for the idea!

Sometimes i experienced the same problem that you in first two photos. Try to render a view from shadow caster view, probably it will help you to see if it is some problem with projection. A fast look to Render Targets in PIX may can help you too.

I solved my problem with a couple of things:
- Maximizing depth precision by placing the view camera as close as possible to the terrain and adjusting its far clip (I couldn't make it work with on ortho camera. The minimal frustum algorithm I found didn't suit my kind of terrain.)
- Making the terrain "full", the first terrain was made from a plane and that caused inaccuracies when the sun lit the terrain from below.

What I'm planing to do now is try to implement variance shadow map following this article: http://www.ziggyware.com/readarticle.php?article_id=161