ML
Desks
register: reading…
desk = task. Roboflow holds the pixels and the SAM-assisted masks; Label Studio is the review desk (draft, not running yet); the registry holds the rows. Tags are the key.
sample sets on the desks, one row per revision; items are the images pushed, never the pool. Tags filter here and on the desk (the same tags sit on every pushed image).
the catalogue as registered; a contract is a filter plus a task plus a review plus an outcome
what came back from the desks and the twin, newest first; SAM3 masks appear here once approved into a Roboflow version and exported
mock no training run has been registered yet; the layout follows the workspace RFC (run · dataset revision · config · metrics · candidate vs incumbent)
mock model versions come from training runs; the first real one will be the environment facet net
export runs = frozen dataset revisions handed downstream; today the Roboflow versions and the Label Studio exports