Home / Services / Data annotation
Data annotation & validationData annotation and validation services.
From creating new ground-truth datasets to reviewing model-generated labels, we provide the people, models, management and quality controls required to deliver dependable AI data. Where a suitable model exists, it pre-labels and our trained team corrects; where judgment is needed, people decide.
Computer vision
Image & Video Annotation
Consistent visual training data for detection, segmentation, classification, tracking and event understanding, across single images and long video sequences.
- Typical inputs
- RGB images
- Video and full-motion video (FMV)
- EO/IR imagery
- Satellite and drone imagery
- DICOM medical imaging
- Annotation types
- Bounding boxes
- Rotated boxes
- Polygons
- Semantic & instance segmentation
- Keypoints and pose
- Object tracking & events
- Classification & attributes
- Methods
- Model pre-labels corrected by trained annotators, or labelling from scratch where no model fits. Guidelines are calibrated on representative samples first.
Select a tab to see the same drone frame structured six ways.
COCO · YOLO · Pascal VOCDOTA · angle attributeCOCO polygons · GeoJSONPNG masks · Cityscapes · COCOPolylines · GeoJSON · OpenLABELMOT 1.1 · MOT173D & Sensor Annotation
Structure LiDAR, point-cloud and multi-sensor data for systems that need to perceive objects, movement and space.
- Typical inputs
- LiDAR point clouds
- Multi-sensor sequences
- Existing pre-annotations
- Annotation types
- 3D cuboids
- Pre-annotation correction
- Methods
- Pre-annotations, from your model or from model-assisted labelling in Point AI Studio, corrected by trained annotators and reviewed in 3D in the browser.
Language & audio
Text & Language Data
Produce and review language data for classification, entity extraction, intent, relevance and model-response evaluation.
- Typical inputs
- Text documents and messages
- Model responses to evaluate
- Annotation types
- Text classification
- Entity extraction
- Intent and relevance
- Model-response evaluation
- Methods
- Label definitions and rating scales written for your task, calibrated on samples, then applied with structured review.
Audio & Speech Data
Prepare human-reviewed audio data through segmentation, timestamping, classification and transcription validation.
- Typical inputs
- Speech recordings
- Existing transcripts
- Annotation types
- Segmentation
- Timestamping
- Audio classification
- Transcription validation
- Methods
- Speakers, noise and overlap marked against written rules; transcripts checked line by line against the audio.
- A00:00.0Thanks for calling, how can I help?
- B00:03.4Hi, I’m calling about an order that never arrived.
- –00:08.1[background noise, overlapping speech]
Validation
Data Validation & QA
Review human- or model-generated labels, correct errors and validate outputs against project-specific acceptance criteria.
- Typical inputs
- Labels from your team or another vendor
- Model predictions and pre-labels
- Your guidelines and acceptance criteria
- Annotation types
- Review of existing labels
- Correction of model pre-labels
- Validation against acceptance criteria
- Ongoing quality operations
- Methods
- Each label is kept, corrected, relabelled or added, and errors are recorded by type. Batch status, error types, agreement and raised questions are visible in Point AI Studio’s quality dashboards.
- Delivered as
- Your format, corrected
How quality is held.
Dense scenes, subtle class boundaries, long video sequences, 3D environments and changing conditions demand more than basic labelling. Point AI teams are calibrated to project-specific rules, raise ambiguous cases early and keep interpretation consistent across batches.
Have a challenging data problem?
Tell us what you’re building and what data you have. We’ll suggest a practical way to process, check and deliver it.