> Unfortunately, it's very difficult - if not impossible - to
> calculate for CSG objects and some primitives, and it's pretty
> meaningless for meshes: the second derivative of a mesh is either
> zero or infinite (undefined, really) anywhere on its surface.  A
> CSG of two spheres would have the same problem, except that the
> curvature would be the radius of one of the spheres or undefined.
> The paper I have here goes into excruciating detail on that problem.

Since every intersection has a normal calculation, what if you were to do a
prepass step and store all the normals for the intersections within the
image, then when you render estimate the curvature? I can see that there
might be problems around the edges of the surface, but could this work
somewhat?

-Mike

Post a reply to this message