Rethinking Out-of-Distribution Detection From a Human-Centric Perspective

TL;DR

This paper advocates for a human-centric approach to out-of-distribution detection, emphasizing the importance of aligning model rejection with human expectations and highlighting limitations of current evaluation metrics.

Contribution

It introduces a human-centric evaluation framework for OOD detection and demonstrates that simple baseline methods can outperform recent complex techniques.

Findings

Simple baseline OOD detection methods perform comparably or better than recent methods.

Model selection significantly impacts OOD detection performance.

Current evaluation metrics may overestimate the progress in OOD detection.

Abstract

Out-Of-Distribution (OOD) detection has received broad attention over the years, aiming to ensure the reliability and safety of deep neural networks (DNNs) in real-world scenarios by rejecting incorrect predictions. However, we notice a discrepancy between the conventional evaluation vs. the essential purpose of OOD detection. On the one hand, the conventional evaluation exclusively considers risks caused by label-space distribution shifts while ignoring the risks from input-space distribution shifts. On the other hand, the conventional evaluation reward detection methods for not rejecting the misclassified image in the validation dataset. However, the misclassified image can also cause risks and should be rejected. We appeal to rethink OOD detection from a human-centric perspective, that a proper detection method should reject the case that the deep model's prediction mismatches the human expectations and adopt the case that the deep model's prediction meets the human expectations. We propose a human-centric evaluation and conduct extensive experiments on 45 classifiers and 8 test datasets. We find that the simple baseline OOD detection method can achieve comparable and even better performance than the recently proposed methods, which means that the development in OOD detection in the past years may be overestimated. Additionally, our experiments demonstrate that model selection is non-trivial for OOD detection and should be considered as an integral of the proposed method, which differs from the claim in existing works that proposed methods are universal across different models.