quote:Original post by Soiled
It would be very interesting to see a comparison of TSM with fixed-up PSM. I''ve asked the authors to contribute to this discussion and post a link to the TSM pdf (which AFAIK makes a comparison with the original PSM). I''ve discussed, with the authors, what I see as a potential problem with TSM in some situations with using the vertex shader (it''s the perspective-correction issue mentioned on page 1 of this thread).

What did they say about perspective-correction issue?
Changing that ''80% rule'' to something like 95% gives even more perspective correction.

However quality of the TSM, like PSM, converges to the regular shadow map quality as the view direction gets more parallel to the light direction.

So it''s not really a generic solution, just a special case one, like the PSM?

I''ve been thinking a lot about this problem, and I think that the only true way of solving this would be to separate shadow map to 3-4 subsets and then apply the TSM?

quote:Original post by Filip Strugar
What did they say about perspective-correction issue?
Changing that ''80% rule'' to something like 95% gives even more perspective correction.

I should probably let the TSM authors answer that question.

Sorry, the perspective-correction I was referring to is perspective-correct interpolation of shadowmap texture coordinates where the texture coords are calculated by interpolating s''=(s/w) and q''=(q/w) between vertices and calculating s''/q'' per pixel. (s,t,r,q) are tex coords passed to vertex and (x,y,z,w) is position of vertex. For perspective-correct shadowmap interpolation (s,t,r,q) are expected to be a linear combination of (x,y,z,w) such as an arbitrary 4x4 matrix multiply.

quote:
I''ve been thinking a lot about this problem, and I think that the only true way of solving this would be to separate shadow map to 3-4 subsets and then apply the TSM?

When I get around to it I''ll probably try TSM in the near distance and blend to one or two hierarchical uniform shadowmaps (see Jon Blow) in the distance. This way only a small amount of geometry around viewer is drawn into shadowmap every frame - the uniform shadow maps only get updated when viewer moves significantly (ie every Nth frame).

Could anyone send me tsm paper, PLEASE?!!! I wrote to authors but got no reply... Pretty, please? fdooch at hotmail.com

I also wrote a short message to the author, and didn''t get any reply yet. Poor guy, he must probably be drowning in requests right now

AFAIK the paper was available on the net for a short while, but the author took it down for some reason. I heard it was supposed to be released for the GDC, but apparently it wasn''t. I suspect the thesis will be presented on Siggraph in may.

Anyway, I''ll just chime in - I''m too curious about this technique Would someone be so nice and send me a copy on heiner at vtales dot com ? I hope that doesn''t classify as copyright infringement. Well, it''s for educational purposes only.