Metric Learning for Novelty and Anomaly Detection

TL;DR

This paper proposes a metric learning approach for detecting out-of-distribution images, including novelty and anomaly detection, demonstrating improved performance over traditional cross-entropy methods in various applications.

Contribution

It introduces a metric learning method for out-of-distribution detection that avoids the limitations of softmax-based classifiers, with extensive experimental validation.

Findings

Achieves comparable or better detection results than previous methods.

Effective in real-world applications like traffic sign recognition.

Distinguishes between related novelty and unrelated anomalies.

Abstract

When neural networks process images which do not resemble the distribution seen during training, so called out-of-distribution images, they often make wrong predictions, and do so too confidently. The capability to detect out-of-distribution images is therefore crucial for many real-world applications. We divide out-of-distribution detection between novelty detection ---images of classes which are not in the training set but are related to those---, and anomaly detection ---images with classes which are unrelated to the training set. By related we mean they contain the same type of objects, like digits in MNIST and SVHN. Most existing work has focused on anomaly detection, and has addressed this problem considering networks trained with the cross-entropy loss. Differently from them, we propose to use metric learning which does not have the drawback of the softmax layer (inherent to cross-entropy methods), which forces the network to divide its prediction power over the learned classes. We perform extensive experiments and evaluate both novelty and anomaly detection, even in a relevant application such as traffic sign recognition, obtaining comparable or better results than previous works.