[Reproducibility Report] Explainable Deep One-Class Classification

TL;DR

This paper reproduces and analyzes the FCDD method for image anomaly detection, confirming its competitive performance and providing insights into its training dynamics and resource requirements.

Contribution

It reproduces key results of FCDD, evaluates its runtime and training performance, and introduces a new analysis methodology using critical difference diagrams.

Findings

FCDD achieves state-of-the-art pixel-wise anomaly detection on MVTec-AD.

Reproduction confirms comparable results on Fashion-MNIST and CIFAR-10.

Analysis reveals training performance dynamics and resource needs.

Abstract

Fully Convolutional Data Description (FCDD), an explainable version of the Hypersphere Classifier (HSC), directly addresses image anomaly detection (AD) and pixel-wise AD without any post-hoc explainer methods. The authors claim that FCDD achieves results comparable with the state-of-the-art in sample-wise AD on Fashion-MNIST and CIFAR-10 and exceeds the state-of-the-art on the pixel-wise task on MVTec-AD. We reproduced the main results of the paper using the author's code with minor changes and provide runtime requirements to achieve if (CPU memory, GPU memory, and training time). We propose another analysis methodology using a critical difference diagram, and further investigate the test performance of the model during the training phase.