Quote:
Originally Posted by TGEN
Like Dirac/Schroedinger?
I'm not sure if n-gons have any tangible benefit over squares, other than avoiding square macroblocks. Octrees might be useful for some scenes (time being the third dimension), but not in the general case.
|
Well, perhaps I should have spoken more carefully and said "non square block". Dirac doesn't use macroblocks but it does use "superblocks" and still does all its motion prediction based on minimum 4x4 block size.
n-gons (dealt internally triangulated, I think) could theoretically give much better intra compression because they aren't limited by vertical and horizontal boundaries... Furthermore, with triangles you could have 2 out of the 3 side's worth of data to interpolate from compared to square block's 2/4 max.
But encoding would be incredibly intensive.
Rather than just a mode decision, you'd have to somehow find optimal tesselations...