Basically, I am trying to test out the Directional Occlusion technique outlined in "The Brutal Art of Brutal Legend" (pg. 75-79). Unfortunately, this necessitates my having a model with DO information baked into vertex data or a texture. So that brings me to my question. As I understand it, I will basically be baking data similar to what is used in a Self-Shadowed Bump map, but raycasting against geometry rather than a heightmap. Are there any tools that will generate this kind of data for me? Can I easily modify a tool to generate it for me? Will I have to make my own custom content processing tool to do this? Any help would be most appreciated. EDIT: The link is just a presentation with one page of short description and three picture pages. Read it, before attempting to answer my question. Also, I am not asking for alternative implementations, I am asking for a means of generating data to test this specific implementation. [Edited by - n00body on March 5, 2010 6:46:47 PM]

I didn't read the article, but it sounds like ambient occlusion including the occlusion vector and not only the ambient factor. Take a look at xnormal.

It is also possible to use spherical harmonics functions to encode the occlusion sphere in vertex or texture data. If you do this, it isn't a big leap to encode statistic effects of radiosity transfer (indirect lighting) and subsurface scattering in there too.

Sorry about being a bit touchy. I made the mistake of trying to work through this problem while being sick and cranky. That's no excuse, but that's how it is.

I am just working off a hunch on their implementation, based on some prior work I did playing around with SS Bump maps. In Valve's implementation, they dynamically convert lights to intensities from three directions, and multiply those with occlusion for each direction, before combining them. This Directional Occlusion information is stored in an SS Bump map, which was generated from a height map.

I am thinking that Double fine's implementation is similar, but generates the DO information based on geometry. My assumption is that they store this data separately, and at a lower frequency, than the bump map data to cut back on storage costs. SS Bump maps are terrible when it comes to compression, and not so good for detail normal mapping either. So I figured they decided to go with a hybrid approach, to get a little of the benefits of SS Bump mapping and tangent-space normal mapping.

This is the notion I want to test, and the original reason I asked my question. I will definitely look into the renderman tools you mentioned, as they sound like the most promising lead thus far. Ultimately though, I would have to write my own tool if I decided to seriously pursue this tech.

Thanks for the help and sorry for being a bit snippy. Take care.