Possible Speed Ups

[william-silversmith]

The code is much faster than it was at the start. An est. 25 core-years is now reduced to 30 minutes for a dense run on connectomics.npy.ckl. However, this is still much longer than it takes to process skeletons with Kimimaro!

One possibility for moving faster is instead of processing single voxels at a time, work with larger and larger cubes and compute the area that way. However, it will take some effort to efficiently distinguish between area that are "safe" and areas that requires more precision.

One other possibility is to determine a bounding box, compute the area for that box, and work on the outside and subtract area from the result. If the box is tight and there are no holes in the interior, then this could be efficient... but it only works if you know there's a single connected component.