Spherical Scale Mapping

very cool!

could you keep optimum resolution using a box primitive instead of a sphere (and AABB instead of bounding sphere, of course)? Then you could get 1 vertex per pixel in the cubemap, and LOD type stuff would be very easy.

It's worth noting that the skull that you use to demo is not, strictly speaking, a "convex" model. What with the eyes being concave and all.

However, there's no overlap from the point of view of the sphere texel that scans that portion of the model, which is why it works.

So it doesn't fail for ALL concave models, only those in which there's an overlap from the point of view of a sphere texel.

Hasn't this been done before?
Displacement mapping?

As Drillian pointed out, this technique does work for some concave meshes, but an important point is that it only works for genus 0 surfaces. It could be interesting to try to extend it to higher genera by using multiple cubemaps or something, but I can't really see a straightforward way of doing it offhand.

Quote:

Original post by aaron_ds Hasn't this been done before? Displacement mapping?

You'd use it with displacement mapping, sure, but this is talking about a particular application of it - namely converting a mesh into a map.