What you get
- Object tracking with persistent IDs
- Action and activity segments with timestamps
- Hand-object interaction tags for egocentric clips
- Keyframe or every-frame labelling
- Clip-level summaries and captions
How we check quality
- ID switches checked through each clip
- Interpolated frames reviewed by a person
- Segment boundaries checked frame by frame
- Clip-level samples re-labelled by QA
How it works
A clear path from first conversation to delivered data.
- GuidelinesWe agree classes, actions and frame rules.
- PilotA few clips are labelled free for your review.
- LabelAnnotators track and tag clip by clip.
- Review and deliverReviewers watch every clip in full before QA.
Common questions
Do you label every frame?
Either every frame or keyframes with checked interpolation, depending on your needs and budget.
Can you label long videos?
Yes. We split long videos into clips with overlap so IDs stay consistent.
Which formats do you deliver?
MOT, CVAT video XML, COCO video JSON, CSV or JSON timestamps.
Related
Start with a free pilot
Send a small sample and your guidelines. No cost, no commitment.