cellmap_flow.dashboard.routes.finetune.yaml_crops

Endpoint for bulk-loading externally annotated crops via a YAML manifest.

Design

A YAML manifest is conceptually a different way to seed an annotation volume, alongside “New Volume” (empty) and “Resume Existing Volume” (copy a prior session). Importing crops writes them straight into the session’s annotation_volume.zarr at their correct physical offsets, so the result is identical in shape to a painted volume — one editable layer in neuroglancer, served via MinIO, picked up by the existing periodic-sync machinery, and consumed by training via VirtualPatchDataset.

Painted scribbles + imported GT crops therefore share one source of truth (the volume zarr). The user can paint over imports to fix GT errors or to add corrections in regions the GT doesn’t cover. The trainer sees the union by construction.

Attributes

logger

Functions

`load_crops_from_yaml_response`(data)	Import crops from a YAML manifest into the session's annotation_volume.
`get_load_crops_progress_response`(load_id)	Return current progress for an in-flight `/api/finetune/load-crops` call.
`read_yaml_file_response`(path)	Return the contents of a YAML file so the dashboard can preview/edit it.

Module Contents

cellmap_flow.dashboard.routes.finetune.yaml_crops.logger

cellmap_flow.dashboard.routes.finetune.yaml_crops.load_crops_from_yaml_response(data)

Import crops from a YAML manifest into the session’s annotation_volume.

Request JSON:

model_name: required
output_path: optional, base path for the session corrections dir
yaml: required, YAML text (or path to a YAML file)
load_id: optional UUID for live progress polling

cellmap_flow.dashboard.routes.finetune.yaml_crops.get_load_crops_progress_response(load_id): Return current progress for an in-flight /api/finetune/load-crops call.

cellmap_flow.dashboard.routes.finetune.yaml_crops.read_yaml_file_response(path): Return the contents of a YAML file so the dashboard can preview/edit it.