Training data, checked twice before your model sees it.
VettedHive is a data labeling team in Bangladesh. We annotate images, video, LiDAR, text, audio and LLM data, and every single item is reviewed by a second person before it ships.
- Data types
- 6
- Reviews per item
- 2
- Pilot cost
- Free
Trusted by data teams at
What we label
Pick a data type to see the tasks we handle and the formats we deliver. If your task isn't listed, send us a sample; most custom schemas are a variation of these.
Boxes, polygons, masks and keypoints drawn tight to the pixel, on any class list you give us.
- 2D bounding boxesTight, axis-aligned or rotated boxes with occlusion and truncation flags.
- Polygons & instance masksVertex-precise outlines for irregular objects, crops, lesions and parts.
- Semantic segmentationEvery pixel assigned a class, with consistent boundaries between neighbours.
- Keypoints & skeletonsBody pose, face landmarks, livestock and custom skeletons with visibility states.
- Classification & taggingSingle- and multi-label image tags, attributes and quality filters.
- OCR & document fieldsText transcription and key–value extraction from receipts, IDs and forms.
Delivered as COCO JSON, YOLO txt, Pascal VOC XML, CVAT XML, Mask PNG, CSV, or your own schema.
Read more about image annotation
Common uses: Retail shelf audits, Crop and weed detection, Medical imaging, Damage assessment.
Built for your industry
Real images, labeled the way we deliver them. Pick an industry to see a full example.
Autonomous driving
Road scenes, LiDAR sweeps, lanes and traffic signs.
Agriculture
Crops, weeds, fruit counts, livestock and drone imagery.
Retail & e-commerce
Shelf audits, product tagging and catalogue attributes.
Healthcare
Medical image regions and clinical text, under strict access rules.
Robotics & drones
Grasp points, obstacles and aerial inspection.
Generative AI
Preference data, evaluations and Bengali language corpora.
How a project runs
Five stages, the same every time, so you always know where your data is and who has touched it.
- 1
Scope
You share sample data and your label guide. We turn it into a written spec with edge cases, class definitions and acceptance criteria.
- 2
Free pilot
We label a small sample at no cost so you can check our work against your own ground truth before committing.
- 3
Label
A trained team works through your data in batches, using your tool or ours, with questions logged and answered in one shared sheet.
- 4
Vet
Every item gets a second pair of eyes. A lead then audits a random sample from each batch and sends back anything below target.
- 5
Deliver
Files arrive in your format with a batch report: accuracy, rework rate and open questions. Corrections after delivery are free.
Why “vetted” is in our name
Cheap labels are fast until you retrain on their mistakes. Our quality loop is built so that no item reaches you having been seen by only one person.
- Written guidelines with pictured edge cases, agreed with you before work starts.
- Accuracy, rework rate and reviewer agreement reported for every batch.
- Repeated errors trigger retraining for the annotator, not just a fix to the file.
Women empowerment
A careful team, powered by women.
VettedHive was founded in 2025 in Bhola, Bangladesh, by annotators who had spent years labeling data for global AI companies. Most of our team are women working from their home district, trained in-house on real projects, with paid training and a clear path from annotator to reviewer to team lead.
Being small means you talk directly to the people labeling your data, and questions get answered in hours, not ticket queues. Meet the team
Send us 100 images. We’ll label them free.
Compare our pilot against your ground truth. If it isn’t good enough, you’ve lost nothing.