Skip to content
Knotest Learn
Join waitlist

Label and correct samples

Publish validated annotations without replacing earlier evidence.

The labelling studio keeps draft annotation work with the same Dataset identity. Earlier published versions and results remain available.

  1. Start from the Dataset

    Use Start labelling for raw images, Continue labelling for an autosaved draft, or Edit labels to correct a saved version.

  2. Work with the appropriate annotation

    The studio supports boxes, editable polygons, and pixel masks with a brush and eraser. Undo/Redo applies to brush strokes as well as other edits. A second reference image can support comparison without opening another editing session.

    Check class assignment and instance boundaries. Empty annotations and missing annotation work have different meanings; retain that distinction when reviewing samples.

  3. Save a version

    Choose Save version to validate and publish labels. The Dataset keeps its ID, name, and folder. Versions containing polygons or masks publish as segmentation Datasets.

Export with the conversion visible

COCO export retains exact masks and holes. YOLO retains polygons and asks before converting masks to bounding polygons.

Use the selected label version in scoring. Compatible corrected Ground Truth can be scored against existing predictions without generating new outputs.