5 replies to this topic

[Ashaman73]

I think about testing out low resolution shadow maps (ie 64x64) for some of my light sources. Instead of sharp shadows I just want some kind of vague, GI like shadows to prevent heavy artifacts, that you would expect from such low resolutions. All light sources will have limited range and I want to sample a texture multiple time and use dithering and a bi-lateral gauss filter to smooth it out along with SSAO.

Have anybody experiences or articles about using low res shadow maps ? Is it worth the effort or will the artifacts just be too strong ?

[Ashaman73]

Argg... wrong forum.. could someone move it to graphics programming and theory..

[Lightness1024]

you could try variance shadow maps ? soft shadows are heavy research

[Ashaman73]

you could try variance shadow maps ? soft shadows are heavy research

My goal is not excactly soft shadows, it is more like a vague, global illumination technique to hint some darker areas behind larger obstancles. By keeping the resolution low, I would update and cache a shadow map for certain light sources. Here's my basic idea sofar:

Shadowmap generation (simple version):
1. Render shadow map of lightsource at higher resolution (512x512).
2. Scale it down (use max depth) to 64x64.
3. Cache it, keep it static.

For each shadow map render a frustum with the projection of the shadow map on the g-buffer.
1. Take multiple samples.
2. The ratio between hits and misses will determine the shadow value (none linear?).

This additional with the SSAO result will be bi-lateral blurred. The difference is, that it is a light independend shading similar to AO. Correct shadow mapping would restrict the incoming light of the associated light source only, my approach would darken the incoming light of all light sources.

Edited by Ashaman73, 18 October 2012 - 03:33 AM.

[MJP]

I've done some experimentation, mostly in an attempt to fake shadows from large area light sources. It mostly works, but I had a lot of problems avoiding biasing artifacts. If you use a large PCF kernel (I was using 7x7) the size of that kernel in world space can be huge, which throws off most biasing techniques. (E)VSM is probably a more promising approach, but I think you'd still run into issues.

[Ashaman73]

I've done some experimentation, mostly in an attempt to fake shadows from large area light sources. It mostly works, but I had a lot of problems avoiding biasing artifacts

This is a good hint. Have you tried to use a hi-resolution shadow map which will be scaled down (max-operator) + bais to avoid it ? The basic idea is, to calculate 1 shadow of the static environment per frame, cache it and fade the shadow map in. If the shadowing effect is not too strong, then this could work.

If you use a large PCF kernel (I was using 7x7) the size of that kernel in world space can be huge, which throws off most biasing techniques.

Hmm... as alternative I could think about a bi-lateral dithering pass before blurring it, just to break up the low res pattern.

but I think you'd still run into issues.

Yes, I'm sure that this will happen, but I'm curious. Maybe I find a quick way to test it out.

Edited by Ashaman73, 18 October 2012 - 11:31 PM.