neu_lib.frame

Voxel indices to physical nanometres (Frame), and the zyx-to-xyz flip.

A level’s voxel size comes from the source’s metadata, never from 2 ** level: real pyramids are anisotropic (halving x and y but not z is common). Arrays are zyx; to_xyz() is the one conversion to a renderer’s xyz, and getting the order wrong mirrors data through the z=x diagonal without any error.

Classes

Frame

Maps voxel indices at one level to physical nanometres, zyx, per axis:

Functions

to_xyz(points_zyx) → numpy.ndarray

(N, 3) zyx points as xyz, in a new C-contiguous float32 array (what a GPU

Module Contents

class neu_lib.frame.Frame

Maps voxel indices at one level to physical nanometres, zyx, per axis: nm = index * voxel_size_nm + origin_nm.

origin_nm is where voxel (0, 0, 0) sits, which is how a crop stays in its parent’s coordinates.

classmethod identity() → Frame

One nm per voxel at the origin — for data already in nm.

to_nm(points_zyx: Sequence | numpy.ndarray) → numpy.ndarray

Voxel indices → nm. Accepts a single point or an (N, 3) array.

to_voxel(points_nm: Sequence | numpy.ndarray) → numpy.ndarray

nm → fractional voxel indices, unrounded.

voxel_box(bounds_nm, *, outward: bool = True) → neu_lib.grid.BBox

A physical (lo, hi) in nm, as a box in this frame’s voxels. Use it to find the same place at another level, or in a crop’s parent.

outward=True floors lo and ceils hi, so the result contains the physical box even when this frame is coarser (a 200-voxel box at 8 nm covers 25 or 26 voxels at 64 nm, depending on where it starts). outward=False rounds each end to the nearest voxel instead.

property is_isotropic: bool

origin_nm: tuple[float, float, float] = (0.0, 0.0, 0.0)

voxel_size_nm: tuple[float, float, float]

neu_lib.frame.to_xyz(points_zyx: numpy.ndarray) → numpy.ndarray

(N, 3) zyx points as xyz, in a new C-contiguous float32 array (what a GPU buffer needs; a reversed view is not contiguous).