cellmap_flow.finetune.session
A finetune session on disk: its annotation volumes, manifest and MinIO sync.
A session is <base>/<YYYYmmdd_HHMMSS>/. Its corrections/ holds the
annotation volumes and _virtual_sources.json, the manifest the trainer
reads; good_regions.json sits beside corrections/.
manifest: the manifest, good regions, and whether a volume is painted.volume: planning, creating, reading and writing annotation volumes.store: which session a base path means, and the volume registry.minio: starting MinIO and serving volumes through it.sync: pulling painted chunks back from MinIO, and the periodic sync.instance: instance corrections, seeded from a segmentation.fill: labelling a whole box of a volume at once, unpainted voxels only.
The submodules import neither flask, neuroglancer, torch nor
cellmap_flow.globals, so the CLI, the trainer and scripts can use them
without a dashboard. State such as the volume registry and MinIO’s is passed
in (the dashboard passes its get_session()’s), never held here. Import the submodule you need; this package imports
none of them itself.