Pixel-accurate image annotation, verified by humans in the EEA
Bounding boxes through panoptic masks, reported in the metrics your team already gates on.
- Human QA to the agreed plan
- EEA-resident (Norway AS)
- 210,000+ contributors
- 30-day erasure SLA
Named-regulator posture, EEA-resident, DPA always
THE HARD 20%
Occlusion, tiny objects, thin boundaries, ambiguous classes
Every tile is annotation geometry on the case that breaks crowdsourced pipelines. One is honestly marked as a gap.
GEOMETRY TIERS
From bounding boxes to panoptic masks, the primitive your task needs
DELIVERY MODEL
Vetted humans, one accountable lead
Annotation is done by people, not a marketplace. Every project runs through the same three roles.
PROJECT-SPECIFIC
- EEA-RESIDENT
- BBOX L2
- POLYGON L2
- MASK L3
LEVELS AND GATES DEFINED PER TASK
- PROJECT LEAD
- QUALIFIED POOL (210,000+ NETWORK)
- QA GATE
- YOUR DELIVERY
- Ramp to production
- PROJECT PLAN
- First delivery gate
- TASK-SET
- Batch cycle
- CONTRACT TARGET
- Staffing model
- VOLUME-SET
The lead who scopes your pilot owns the specification, gates, and report. Ramp and capacity are defined by contract.
Quality you can audit: kappa-gated annotators, gold-set benchmarking, error taxonomy
Quality drift from annotator churn and fatigue is the real failure mode on long projects. We run a living QA system, not a one-off pass, and the team labeling your data is a named cohort, not an anonymous crowd.
Agreement method
TASK-SPECIFIC AGREEMENT, CALIBRATION-SET GATED
Metric, reference set, and threshold are defined for the engagement.
Human QA plan
Acceptance evidence
Erasure SLA
The agreement method, reference set, sample, and threshold are selected for the task.
Confusion matrix
Per-class agreement; diagonal dominance shows where the guideline needs tightening.
Error taxonomy
| Error type | Critical | Major | Minor |
|---|---|---|---|
| Boundary | route | review | sample |
| Class | route | review | sample |
| Attribute | route | review | sample |
| Miss | route | review | sample |
Boundary, class, attribute, miss; classified by critical, major, minor severity.
- 01
Gold-set ground truth
Held-out expert-labeled frames seeded as honeypots so every annotator is measured against known-correct ground truth continuously.
- 02
Consensus multi-pass
Multiple annotators on the same frames, keeping elements at a documented agreement threshold, with kappa tracked per class.
- 03
Audit sampling
Statistical audit sampling with acceptance thresholds set tighter for safety-critical and ambiguous classes.
- 04
Drift control
Continuous sampling and dashboards across the project, with annotator retraining when class-level agreement slips.
Four gates, one named cohort, quality that holds from the first thousand images to the ten-millionth.
This quality model, measured on your data against your acceptance criteria.
METRIC GATES / BY VERTICAL
The metric your industry gates on
This matrix shows common metric families, not YPAI performance. Metric selection and thresholds are set for the actual task and reported against the contract.
- IoU, drawn as the overlap it is
- PQ: the segmented whole
- kappa: annotator agreement
| IoU | COCO mAP | PQ | boundary-F1 | agreement | |
|---|---|---|---|---|---|
| Automotive ↗ | PRIMARY | TASK-SET | TASK-SET | PRIMARY | TASK-SET |
| Medical imaging | PRIMARY | OPTIONAL | TASK-SET | PRIMARY | TASK-SET |
| Retail | TASK-SET | PRIMARY | OPTIONAL | TASK-SET | TASK-SET |
| Robotics | PRIMARY | TASK-SET | PRIMARY | PRIMARY | TASK-SET |
| Geospatial | PRIMARY | TASK-SET | PRIMARY | TASK-SET | TASK-SET |
ARTICLE 10 / DELIVERED
The dossier ships with the dataset
Not a policy PDF. The audit answers, pre-written.
EEA-RESIDENT
Your image data stays in the EEA, pinned at scoping, with architecture your security team can review
US CLOUD Act exposure means even a US vendor with an EU subsidiary can be compelled to hand over data. We are a Norwegian company processing on self-hosted European servers, under Norwegian jurisdiction with per-project transfer controls.
Formats, model-assist, residency, consent, erasure
- Formats What export formats do you deliver?
- COCO (including panoptic), YOLO, Pascal VOC, CVAT, PNG masks, and custom JSON or your schema. Customer-owned work product, no reuse rights retained.
- Model-assist If you use model-assist, who is accountable for the label?
- Model labels are suggestions only. Human verification follows the agreed acceptance and sampling plan, edge cases and safety-critical classes can receive double-blind review, and every change is logged so a human stays accountable for the delivered geometry.
- Residency Where is our image data processed?
- EEA-resident. Norwegian company, self-hosted European servers, EEA contributor network, under Norwegian jurisdiction with per-project transfer controls. No non-EEA support access without your written instruction.
- Consent How do you handle consent and special-category image data?
- We are your processor and document your Article 6 and, where relevant, Article 9 basis. Faces and plates are blurred in-tool before annotators see them, exports are pseudonymized, and consent withdrawal is supported.
- Erasure What is your erasure commitment?
- A 30-day erasure SLA across systems for image and annotation deletion, with a signed DPA on every engagement and a sub-processor list with change notifications.
Scope a metric-reported image annotation pilot
We return a sample set with the IoU, AP, or kappa report you specify, so you verify quality before you scale.
- Metric-reported pilot
- Measured in IoU and COCO AP
If image annotation is not the right fit for your project, we will say so directly.
One business day reply. EEA-resident, Norway.
IMAGE · VIDEO · AUDIO · LIDAR · TEXT · EVAL / ONE GOVERNED PIPELINE