Raising the Bar on the Evaluation of Out-of-Distribution Detection

TL;DR

This paper introduces a new framework for evaluating out-of-distribution detection methods by generating challenging OoD samples that are perceptually or semantically similar to in-distribution data, revealing limitations of current methods.

Contribution

It defines two categories of OoD data, proposes a GAN-based method to generate such samples, and demonstrates that existing detection methods are less effective against these new benchmarks.

Findings

State-of-the-art OoD methods are less robust on the new benchmarks.

Models performing well on new benchmarks also perform well on traditional ones.

The new benchmarks can evaluate OoD detection without separate OoD datasets.

Abstract

In image classification, a lot of development has happened in detecting out-of-distribution (OoD) data. However, most OoD detection methods are evaluated on a standard set of datasets, arbitrarily different from training data. There is no clear definition of what forms a ``good" OoD dataset. Furthermore, the state-of-the-art OoD detection methods already achieve near perfect results on these standard benchmarks. In this paper, we define 2 categories of OoD data using the subtly different concepts of perceptual/visual and semantic similarity to in-distribution (iD) data. We define Near OoD samples as perceptually similar but semantically different from iD samples, and Shifted samples as points which are visually different but semantically akin to iD data. We then propose a GAN based framework for generating OoD samples from each of these 2 categories, given an iD dataset. Through extensive experiments on MNIST, CIFAR-10/100 and ImageNet, we show that a) state-of-the-art OoD detection methods which perform exceedingly well on conventional benchmarks are significantly less robust to our proposed benchmark. Moreover, b) models performing well on our setup also perform well on conventional real-world OoD detection benchmarks and vice versa, thereby indicating that one might not even need a separate OoD set, to reliably evaluate performance in OoD detection.