posted May 03, 2000 01:23 PM

A new topic for discussion...I've read and reviewed the work of Jonathan Blow's Hierarchical Split Decision method and would like to create an algorithm from it. The paper can be found here:

http://www.bolt-action.com/dl_papers.html

The paper talks about using an isosurface (in their case, a sphere) which contains other spheres hierarchicaly [I hope I'm spelling that right]. I'm curious as to how these sphere radii are determined, and how this differs from the one-dimentional transformation that they were trying to replace.

In other words, they take 3D eye coordinate and a sphere radius (a technically 1D entity) and say that this is much better than using a 'variance' scheme which takes a 3D eye coordinate and a 1D error value.

I'm working on applying this technique to a simplified quadtree scheme, and I think it would be very efficient. Each QuadTree node would contain a sphere radius that is the maximum of its children plus some detail factor. When the eye crossed into the sphere, the node would be split. When it left the sphere, the node would be merged.

Any ideas on this scheme (it's a good paper, do take a look). The texture cache is also very interesting... but one step at a time.