Unsupervised evaluation for out-of-distribution detection

Yuhang Zhang, Jiani Hu, Dongchao Wen, Weihong Deng

Abstract: Highlights•We propose unsupervised evaluation of out-of-distribution (OOD) detection methods.•We introduce a benchmark of 200 different OOD datasets.•We find a nearly linear relationship between our proposed unsupervised indicator and OOD detection performance.•Our method achieves state-of-the-art performance and generalizes well under different OOD detection methods, ID/OOD datasets, dataset sizes, and ID:OOD ratios.