Data labeling
Video annotation and object tracking that holds across every frame
Consistent object IDs through occlusion, re-entry and camera motion, plus action, event and temporal labels. Our annotators track what matters and reviewers check continuity, not just single frames.
- Pilot phase
- Two or more review layers
- You own every label
A label in frame 1 has to mean the same thing in frame 900.
Video adds a dimension most labeling errors hide in: time. An ID that swaps when two pedestrians cross, a box that drifts during interpolation, or an action tagged three frames late will quietly teach your tracker the wrong thing.
We annotate keyframes by hand, interpolate between them and then review the full sequence at playback speed, checking that IDs persist, boxes stay tight and events start and stop on the right frame.
Used for
- Dashcam and robotics perception
- Sports and fitness analytics
- Retail footfall and shelf interaction
- Security and safety event detection
- Drone and inspection footage
Industries: Autonomous vehicles · Manufacturing & robotics · Retail & e-commerce
What our video annotation covers
Combine any of these in one workflow and one guideline.
Multi-object tracking
Boxes or polygons with persistent IDs across frames, through occlusion and re-entry.
Action & event tagging
Start and end frames for actions, gestures, incidents and custom events.
Temporal segmentation
Scene and phase boundaries for sports, surgical and process video.
Frame classification
Per-frame or per-clip tags for quality, content and scene type.
Keypoint tracking
Pose and landmark tracking for sports analytics, fitness and AR.
Sensor-synced video
Camera streams aligned with LiDAR or telemetry for fusion datasets.
Formats and tools
Delivery formats
- MOT
- COCO video
- CVAT video XML
- JSONL per frame
- Custom schema
Tools we work in
- CVAT
- Encord
- Labelbox
- V7
- SuperAnnotate
- Your in-house tool
Pricing
Video object tracking from $0.050 per object per frame, and from $0.035 at volume. Dense scenes are quoted after a sample.
How we keep it accurate
Every project runs on the same four-check QA process. See how our QA works.
ID-switch audits
Reviewers scrub each sequence for identity swaps and fragmented tracks, the errors that hurt trackers most.
Interpolation checks
Interpolated frames are spot-checked so boxes don’t drift between keyframes.
Event timing tolerance
Agreed frame tolerances for event start and end, measured on every batch.
How a project runs
Scope and guidelines
We review a sample, agree classes, edge cases and acceptance criteria, and sign your NDA.
Pilot phase
Pilot phase helps you judge quality, speed and communication.
Certified team
Annotators pass a gold-set test before they are cleared for production.
Production and reporting
Batches with QA reports, in your format, on an agreed schedule.
Video annotation questions
How do you keep object IDs consistent?
Annotators follow written re-identification rules for occlusion and re-entry, and reviewers watch each sequence at playback speed specifically to catch ID switches.
Do you label every frame?
Usually we label keyframes and interpolate, then review. For fast motion or strict accuracy we label more densely. We agree the approach during the pilot.
Can you handle long sequences?
Yes. Long sequences are split into overlapping chunks with ID hand-off rules, so tracks stay continuous across chunk boundaries.
What frame rates and resolutions can you work with?
Any your tool supports. For very high-resolution or high-frame-rate footage, we agree a sampling rate with you first.
Send a sample. See the labeling quality first.
Share a few files and a line about your goals. A project lead replies within one business day with an approach, a timeline and a price.