Out-of-Distribution Detection Without Class Labels

TL;DR

This paper introduces an unsupervised approach for multi-class out-of-distribution detection that uses clustering to generate pseudo-labels, enabling the application of standard detection methods without requiring class labels during training.

Contribution

The work presents a novel two-stage method combining unsupervised clustering with supervised OOD detection, eliminating the need for manual labels in multi-class OOD detection.

Findings

Achieves competitive performance compared to state-of-the-art methods.

Effectively discovers pseudo-classes that improve OOD detection accuracy.

Provides extensive analysis and ablation studies to validate the approach.

Abstract

Out-of-distribution detection seeks to identify novelties, samples that deviate from the norm. The task has been found to be quite challenging, particularly in the case where the normal data distribution consists of multiple semantic classes (e.g., multiple object categories). To overcome this challenge, current approaches require manual labeling of the normal images provided during training. In this work, we tackle multi-class novelty detection without class labels. Our simple but effective solution consists of two stages: we first discover "pseudo-class" labels using unsupervised clustering. Then using these pseudo-class labels, we are able to use standard supervised out-of-distribution detection methods. We verify the performance of our method by a favorable comparison to the state-of-the-art, and provide extensive analysis and ablations.