cellmap_flow.io.ome

OME-NGFF coordinate conventions.

OME-NGFF places translation at the centre of voxel 0 (Neuroglancer’s ome.ts subtracts half a voxel for the same reason), while everything in cellmap-flow works with the lower corner of voxel 0. Readers convert with ome_corner and writers with ome_translation; nothing else should add or subtract half a voxel. multiscales_attrs writes the attributes.

Attributes

CHANNEL_AXIS_NAMES

Functions

`ome_corner`(→ List[float])	The lower corner of voxel 0, from an OME-NGFF `translation`.
`ome_translation`(→ List[float])	The OME-NGFF `translation` (voxel-0 centre) for a lower `corner`.
`multiscales_attrs`(→ Dict[str, Any])	OME-NGFF `{"multiscales": [...]}` attributes for a pyramid.
`singlescale_attrs`(→ Dict[str, Any])	`multiscales_attrs` for one array `arr_name` whose voxel 0 has its

Module Contents

cellmap_flow.io.ome.ome_corner(translation: Sequence[float], scale: Sequence[float]) → List[float]

The lower corner of voxel 0, from an OME-NGFF translation.

Janelia pyramids store translation = scale/2 - 4 per level, so every level’s corner is -4 nm; read as a corner, each level sat half its voxel off (s1 8 nm, s2 16 nm) and the levels did not even agree with each other. translation and scale must be in the same units.

cellmap_flow.io.ome.ome_translation(corner: Sequence[float], scale: Sequence[float]) → List[float]: The OME-NGFF translation (voxel-0 centre) for a lower corner.

cellmap_flow.io.ome.CHANNEL_AXIS_NAMES = ('c', 'c^', 'channel')

cellmap_flow.io.ome.multiscales_attrs(axes: Sequence[str], units: Sequence[str], levels: Sequence[Tuple[str, Sequence[float], Sequence[float]]], name: str = '', version: str = '0.4') → Dict[str, Any]

OME-NGFF {"multiscales": [...]} attributes for a pyramid.

levels are (path, voxel_size, corner), one entry per axis in axes order, with corner the lower corner of voxel 0. Spatial axes get translation = corner + voxel_size / 2 and the unit from units; the channel axes (named “c”, “c^” or “channel”) are written as they are, with type “channel” and no unit. voxel_size is written as given.

cellmap_flow.io.ome.singlescale_attrs(arr_name: str, voxel_size: Sequence[float], offset: Sequence[float], units: Sequence[str], axes: Sequence[str]) → Dict[str, Any]

multiscales_attrs for one array arr_name whose voxel 0 has its lower corner at offset.

Writing the corner as the translation put every output half a voxel off in Neuroglancer and in any OME reader.