Machine learning annotation quality control

Know your dataset before your model does.

Find label, geometry, split, and image problems before training a machine learning model. Get one clear score and a practical repair list.

Compare two versions

CSV, JSON, JSONL, COCO, YOLO, Pascal VOC, and LabelMe. Up to 150MB.

EXAMPLE QUALITY REPORTPreview
Example score
92
Excellent
Labels100/100
Geometry96/100
Image quality78/100
Illustrative report. Upload a dataset to calculate your score.
Private by design
Files stay in this browser
01

Clean labels

Find naming drift, empty values, duplicates, and imbalance.

02

Trust geometry

Validate boxes, polygons, masks, keypoints, and overlaps.

03

Inspect images

Flag blur, exposure, corruption, rotation, and duplicates.

04

Ship evidence

Export repair lists, reports, policies, and version comparisons.

No cookies. No account. No dataset telemetry. Your files never leave your browser.