Annotation and labelling
Annotation with a QC step that happens before delivery
Image, video and text labelling, plus review of whatever your pre-labelling model already produced. The sampling pass is not optional and not an upsell.
The work
What we take on
If your task isn't listed, ask anyway — the answer is either yes, or a straight no with a reason.
- Image and video bounding boxes, polygons and keypoints
- Classification and content categorisation
- Review and correction of pre-labelled data
- Gold-set creation and inter-annotator agreement checks
- Edge-case triage and taxonomy feedback
How it's delivered
- Tools
- Yours — CVAT, Label Studio, Roboflow, or an internal tool
- Unit of work
- Per image, per frame, per annotation, or per seat-hour
- Quality signal
- Sample pass rate plus inter-annotator agreement where you want it
- Rework
- Counted and reported. Ours to fix when the error is ours.
Honest scoping
When this is a good fit, and when it isn't
Taking work we'd do badly costs us more than turning it down. This is the list we'd give you on the call anyway.
Good fit
- Pre-labelling gets you most of the way and you need humans on the rest
- Your taxonomy has genuine edge cases that need judgement
- You want the same annotators improving on your data over months
Not us
- You need specialist clinical or geospatial expertise
- You want the cheapest possible price per unit — a crowd platform will beat us