What is out-of-distribution (OOD) detection evaluation?

It evaluates model ability to detect unfamiliar inputs.

OOD detection ensures models can identify inputs that differ from training distribution, often using confidence scores or density estimation.