Anomaly Detection with Test Time Augmentation and Consistency Evaluation

TL;DR

This paper introduces TTA-AD, a simple post-hoc anomaly detection method that leverages prediction consistency under test-time augmentations to distinguish in-distribution from out-of-distribution data, achieving high efficiency and performance.

Contribution

The paper proposes TTA-AD, a novel anomaly detection approach based on test-time augmentation and consistency evaluation, with empirical and theoretical validation.

Findings

TTA-AD achieves comparable or better detection performance than existing methods.

TTA-AD reduces running time by 60-90% compared to classifier-based algorithms.

Prediction consistency under augmentation is key to effective anomaly detection.

Abstract

Deep neural networks are known to be vulnerable to unseen data: they may wrongly assign high confidence stcores to out-distribuion samples. Recent works try to solve the problem using representation learning methods and specific metrics. In this paper, we propose a simple, yet effective post-hoc anomaly detection algorithm named Test Time Augmentation Anomaly Detection (TTA-AD), inspired by a novel observation. Specifically, we observe that in-distribution data enjoy more consistent predictions for its original and augmented versions on a trained network than out-distribution data, which separates in-distribution and out-distribution samples. Experiments on various high-resolution image benchmark datasets demonstrate that TTA-AD achieves comparable or better detection performance under dataset-vs-dataset anomaly detection settings with a 60%~90\% running time reduction of existing classifier-based algorithms. We provide empirical verification that the key to TTA-AD lies in the remaining classes between augmented features, which has long been partially ignored by previous works. Additionally, we use RUNS as a surrogate to analyze our algorithm theoretically.