Core AI Data Terms
Active learning
A training strategy where uncertain or high-value examples are routed for human review so annotation effort is focused where it can improve model performance most.
Annotation guideline
A written instruction set that defines labels, edge cases, examples, counterexamples and acceptance rules for reviewers.
Bounding box
A rectangular annotation used to identify object location in images or video. See image annotation services.
Classification
Assigning a category or label to an item such as an image, document, utterance, product or content item.
Computer vision
AI systems that interpret visual data such as images, video, medical scans, retail shelves, roads, products and industrial scenes.
Data curation
Selecting, cleaning, organizing and preparing datasets so they represent the model objective and production environment.
Data labeling
Assigning structured labels to data so a machine learning model can learn from examples.
Dataset validation
Checking training data for label quality, completeness, consistency, class balance and suitability for model training or evaluation.
Ground truth
The validated reference label or answer used to train, test or evaluate an AI model.
Human-in-the-loop
A workflow where human reviewers guide, validate, audit or correct AI outputs. It is central to high-risk, ambiguous and quality-sensitive AI systems.
Instance segmentation
A pixel-level annotation method that separates each object instance in an image, even when objects belong to the same class.
Keypoint annotation
Marking specific points on an object or body, often used for pose estimation, product landmarking and visual measurement tasks.
LiDAR annotation
Labeling 3D point cloud data for autonomous vehicles, robotics, mapping and spatial AI. See LiDAR annotation services.
Named entity recognition
An NLP annotation method that identifies entities such as people, companies, products, dates, locations and domain-specific terms in text.
OCR annotation
Preparing text extraction, document labeling and field validation data for optical character recognition and document AI systems.
Polygon annotation
A visual annotation method that traces object boundaries more precisely than a bounding box.
RLHF
Reinforcement learning from human feedback, where human preferences or evaluations help improve model behavior, especially in LLM workflows.
Semantic segmentation
Pixel-level labeling where each pixel is assigned to a class such as road, vehicle, product, tumor, shelf or background.
Synthetic data
Artificially generated training data used to supplement real datasets, stress-test models or fill data gaps.
Training dataset
The collection of labeled examples used to train or fine-tune a machine learning model.
Video annotation
Labeling objects, actions, events and movements across frames. See video annotation services.