XOOD: Extreme Value Based Out-Of-Distribution Detection For Image Classification

TL;DR

The paper introduces XOOD, a novel extreme value-based framework for out-of-distribution detection in image classification, which outperforms existing methods in accuracy and efficiency by leveraging activation layer signals.

Contribution

It proposes two new algorithms, XOOD-M and XOOD-L, that utilize extreme value signals for OOD detection, with superior performance and faster inference.

Findings

Outperforms state-of-the-art OOD detection methods

Reduces false-positive rate (FPR95) by 50%

Improves inference time by an order of magnitude

Abstract

Detecting out-of-distribution (OOD) data at inference time is crucial for many applications of machine learning. We present XOOD: a novel extreme value-based OOD detection framework for image classification that consists of two algorithms. The first, XOOD-M, is completely unsupervised, while the second XOOD-L is self-supervised. Both algorithms rely on the signals captured by the extreme values of the data in the activation layers of the neural network in order to distinguish between in-distribution and OOD instances. We show experimentally that both XOOD-M and XOOD-L outperform state-of-the-art OOD detection methods on many benchmark data sets in both efficiency and accuracy, reducing false-positive rate (FPR95) by 50%, while improving the inferencing time by an order of magnitude.