Solutions · Annotation & curation
Labels in your schema, reviewed twice.
Language, temporal, 2D/3D and sensor-fusion labels on your corpus or ours. Every rejection reported with a reason code. Nothing substituted quietly.
Label types
What we label
-
Language
Stage-level descriptions of each step of the task, with frame ranges.
-
Temporal
Task stages, segment boundaries and the moments that decide success or failure.
-
2D and 3D
Segmentation, tracking and 3D cuboids.
-
Sensor fusion
Labels kept consistent across cameras, depth and LiDAR.
Curation & formats
Curated, in your schema
- Deduplication
- Near-duplicates removed
- Near-duplicate episodes found, removed and reported.
- Curation
- Selected to your criteria
- Task coverage, quality thresholds and balance across scenes and embodiments.
- Your schema
- COCO · JSONL · KITTI · custom
- We label to your schema and guidelines. Edge-case rules are agreed with you before the first batch.
Quality
Quality you can audit
Who does the work
Trained annotation teams on our production floor in Vientiane, Lao PDR, working to guidelines and QC methods developed at our research lab in Incheon. Every annotator has signed an NDA and passed a background check.
In every batch report
- Two independent reviewers for every episode
- Agreement between reviewers, reported per batch
- Every rejection listed with a reason code
- Review notes: what changed and why
Other stages
Pairs well with
Evaluation batch A
Start with a free evaluation batch.
1,000 of your episodes, in 14 days. Free, and graded before you commit.