I'm not sure I understand how binning applies in e.g. segmentation of point clouds into distinct objects. The data would likely contain a mix of unknown distrib

I'm not sure I understand how binning applies in e.g. segmentation of point clouds into distinct objects. The data would likely contain a mix of unknown distributions, partially observed (due to occlusions) and not easily parametrized (chair, table, toaster, etc.)... Locally you can find planar patches though, so correlation can still be useful.