Technical GlossaryComputer Vision

Open-Set Recognition

An approach that enables a model to flag unseen classes as unknown instead of assigning them an overconfident incorrect label.

Open-set recognition represents a safer classification philosophy that acknowledges real-world distributions are broader than training data. In production, forcing every input into one of the known classes can lead to serious errors. For this reason, the ability to separate unknown examples becomes critical in security, medical triage, and quality-control systems.