cellmap_flow.dashboard.routes.finetune

The finetune tab’s routes, and the viewer-layer routes scripts use.

Importing the package puts every route on finetune_bp, which dashboard.app registers. Each module has its own:

training: submit, restart, the job list, a job’s status and log (and its live stream), cancel and stop-early;
annotation_core: the models to annotate for, a new annotation volume, and the user’s saved settings;
annotation_sessions: listing earlier sessions and resuming one;
yaml_crops: importing crops from a YAML, and reading a YAML file;
overlay: an annotation volume’s layer, the annotated-regions boxes, and syncing the annotations from MinIO;
good_regions: marking the current view as a region the model gets right;
rank_routes: which model to start from: the models ranked by how well each already matches the paint, and the paint carried to the one picked;
channel: the channel of a model of several that its layer shows;
view_labels: labelling the view from the model’s prediction, or all background;
ai_annotate: a hosted model’s first guess at one plane, staged for the user to accept or reject;
instance_correction: instance-correction volumes;
layers: adding, removing and renaming viewer layers;
publish: signing in to Hugging Face, and publishing a model a job finetuned there, under the user’s own account.

Submodules