API reference

Every module of every package, with its classes, functions and docstrings.

This is parsed from the source, never imported, which is worth knowing because it is what makes the reference complete. Half of this suite cannot be installed from PyPI at all — tensorstore, vol2mesh, dvidutils, kimimaro and neuclease are conda-only, and neu-draw needs a GPU stack the documentation build has no use for. An importing reference would have to either stub all of that out or quietly omit whatever failed to import; this one reads the files.

What it is not

A curated public API. Everything is here, private helpers and all, because these packages are read as often as they are called — the reason a function does what it does is usually in its docstring, and that is the thing worth coming here for. For the surface that is meant to be called, start from each package’s README, or from the table below.

The one deliberate omission is the cli modules. Those are documented by the CLI reference, rendered from the real ArgumentParser objects, which is a better page than an API listing of argparse plumbing.

The packages, in dependency order

Nothing lower may import from anything higher, and packages at the same tier do not import each other.

neu_lib

The vocabulary every tier shares. numpy and nothing else.

Start at BBox, Frame, Mesh, Skeleton, ScaleInfo, align_box, skeleton_tube

blockrun

The dask/SLURM substrate. Knows nothing about electron microscopy.

Start at block_map, iter_blocks, Manifest, start_dask

neu_vol

Volume I/O: tensorstore backends, storage profiles, the conversion ops, source metadata.

Start at convert, open_backend, scale_spec, read_scales, describe, location, readback

neu_morpho

Per-body meshes, published into the volume, and measurement of meshes and skeletons.

Start at pipeline, measure, precomputed, occupancy

neu_mark

DVID annotations into tables, and on into neuroglancer.

Start at notebook, tables, explore, rule, segprops, annsource

neu_glance

Everything a remote viewer consumes: states, layers, links, shaders.

Start at state.build_state, layers, shaders, sources, movie

neu_proc

Array processing: filters, label tools and distance transforms. Only the pure array-in-array-out half so far.

Start at dilate, dust, ops.kernels, ops.backend

neu_eval

Comparing a segmentation against a reference labeling: the metrics, and where the two disagree.

Start at contingency, compare, voxel, assign, disagree, adjudicate

neu_draw

Local 3D rendering in a notebook, on pygfx.

Start at show, build_scene, Scene, sources, Legend

Where to start, by what you are doing

you want to	reach for
do box or grid arithmetic	neu_lib.grid — BBox, align_box, lcm_grid
turn voxels into nanometres	neu_lib.frame — Frame, to_xyz
run something per block, resumably	blockrun.engine — block_map, iter_blocks — with blockrun.manifest
open a volume and read a region	neu_vol.backends — open_backend, and scale_spec for the level
ask what a volume is	neu_vol.source_metadata, neu_vol.scales
read or write a store, local or s3://	neu_vol.location
convert or copy a source	neu_vol.ops.convert
read published meshes and skeletons back	neu_vol.readback
measure morphology of published output	neu_morpho.measure
pull DVID annotations into DataFrames	neu_mark.notebook
inspect and parse neuron names	neu_mark.explore, neu_mark.rules
build a neuroglancer state or layer	neu_glance.state, neu_glance.layers
render an animation through a real viewer	neu_glance.movie
draw meshes and skeletons in a notebook	neu_draw.scene — build_scene, Scene — then neu_draw.show

Two conventions that run through all of it

Both fail silently when broken, so they are worth knowing before reading any signature:

• zyx in memory, xyz on disk. Every region argument, Mesh.vertices_zyx and kimimaro’s vertices are zyx. Both precomputed formats store xyz, and the flip happens at the boundary. Getting it wrong mirrors output through the z=x diagonal.

• One model space: physical nanometres, via each level’s real voxel size — never an assumed 2 ** level factor, because real pyramids are anisotropic.

---

Looking for a module by name? The module index lists every one alphabetically, and the search box knows every docstring.