A few blobs are all you need

The model proposes where it is uncertain, a pathologist confirms the tissue class, the adapters retrain, performance is logged — thirty rounds. Switch protocol to see how the annotator's work, and the cost, change.

annotation-station
iteration 0 / 29

Review queue

—
fine-tuning adapters
10 epochs
selected this iteration selected in an earlier iteration current region (on the main stage)
ready

Macro Dice · w/o background

Validation
—
Test
—

Annotation budget

Regions labelled
—
Pixels of dataset
—

Loop stage

1 select
2 annotate
3 train
4 evaluate