Back to experimenting with meshing algorithms, trying to add variable resolution support while keeping the output manifold.
Left image is just directly meshing SDF samples in an octree (complete with red non-manifold edges where resolution changes). Right is a dual method, edges are a mess so it doesn't look manifold, but it is!
It turns out there are two (very different) papers that both call themselves "dual marching cubes" – are you referring to Nielson '04, or Schaefer & Warren '05?
Add comment