Data annotation (Data labeling)

Data annotation, also called data labeling, is the process of assigning meaningful labels or metadata to raw data so it can be used to train, evaluate, or monitor machine learning systems.

It turns unstructured inputs such as images, point clouds, audio, and video into structured, machine-readable training data.

Depending on the task, annotations can take forms such as bounding boxes, semantic masks, keypoints, and classifications.

Data annotation is a core step in supervised learning and strongly affects model performance.