Label (Tag)

A label (tag) is an annotation assigned to a data sample or a specific region within it, indicating its category, attribute, or semantic meaning. Labels are usually created by human annotators following guidelines, though they can also be generated through automated or semi-automated methods.

Depending on the task, labels can be categorical, numeric, textual, or structured, such as bounding box coordinates for detection or pixel masks for segmentation.

In supervised learning, labels define the target outputs the model is trained to predict. Label format and granularity must match the model outputs and evaluation metrics. Errors or inconsistency can degrade both training and evaluation.