cellmap_flow.review_index

Build a review index, the SQLite file the dashboard’s Review tab opens, from an instance segmentation.

python -m cellmap_flow.review_index LABELS OUT.sqlite [–queue NAME …]

LABELS is anything ImageDataInterface reads (zarr v2/v3, N5, precomputed; a multiscale group is read at its finest level), with 0 as background. The index has the schema cellmap_flow.review reads: one instances row per label with its voxel count, bounding box (voxels, half-open) and centroid (voxels, and in world nm), plus one rank_<queue> column per queue; one empty ledger row per label; and meta.

The volume is read in blocks of whole chunks and only per-label sums are kept, so it need not fit in memory.

Queues are chosen with --queue from QUEUES. Other orderings are data: any rank_<name> column added to instances later (say, from per-instance intensity) shows up in the tab as a queue called <name>.

Attributes

`QUEUES`
`DEFAULT_QUEUES`
`MIN_BLOCK_EDGE`
`MAX_LABEL`

Functions

`build_index`(→ Dict[str, object])	Write the review index for `labels_path` to `db_path`.
`main`([argv])

Module Contents

cellmap_flow.review_index.QUEUES

cellmap_flow.review_index.DEFAULT_QUEUES = ('smallest', 'random')

cellmap_flow.review_index.MIN_BLOCK_EDGE = 128

cellmap_flow.review_index.MAX_LABEL = 9223372036854775807

cellmap_flow.review_index.build_index(labels_path: str, db_path: str, queues: Sequence[str] = DEFAULT_QUEUES, seed: int = 0, block_shape: Sequence[int] | None = None) → Dict[str, object]

Write the review index for labels_path to db_path.

Refuses to overwrite db_path: its ledger holds the verdicts. Returns {"n_instances", "voxel_size_nm", "offset_nm"}.

cellmap_flow.review_index.main(argv=None)