Class

A class is a category that a data sample can belong to, defined by the labeling scheme for a task.

In object detection, each detected instance is assigned to a predefined class such as “car,” “pedestrian,” or “traffic sign.”

The set of all possible classes is called the label set. Class definitions should be mutually exclusive and include clear inclusion and exclusion rules in the labeling guidelines to reduce ambiguity and label noise.

The number and distribution of classes in a dataset can strongly affect training dynamics, including issues from class imbalance.