neu_proc¶
neu-proc: array processing for large 3D EM volumes.
scipy and scikit-image filters and the Seung-lab label tools (cc3d, fastremap, fastmorph,
edt, fill_voids), applied to whole arrays. Only neu_proc.ops exists so far: array
in, array out. Running ops blockwise across a volume, with halos and seam reconciliation,
is planned; an op that works on one array is the reference that runner will be checked
against.
Functions take voxel parameters. Converting a physical length into voxels for a given
level happens in neu_proc.ops.kernels, never inside a function: a halo derived from
anything but the converted parameter is wrong in a shell around every seam, which shows
as faint block edges rather than as an error.
Top-level names resolve lazily (PEP 562), so importing the package pays for neither scipy nor dask until a name is used.