DSR -- A dual subspace re-projection network for surface anomaly detection
This addresses the challenge of realistic anomaly synthesis for unsupervised surface anomaly detection, particularly for near-in-distribution cases, with incremental improvements over existing methods.
The paper tackles the problem of detecting near-in-distribution anomalies in surface anomaly detection by proposing DSR, a dual subspace re-projection network that generates anomalies at the feature level without image-level synthesis, achieving state-of-the-art results with improvements of 10% AP in detection and 35% AP in localization on the KSDD2 dataset.
The state-of-the-art in discriminative unsupervised surface anomaly detection relies on external datasets for synthesizing anomaly-augmented training images. Such approaches are prone to failure on near-in-distribution anomalies since these are difficult to be synthesized realistically due to their similarity to anomaly-free regions. We propose an architecture based on quantized feature space representation with dual decoders, DSR, that avoids the image-level anomaly synthesis requirement. Without making any assumptions about the visual properties of anomalies, DSR generates the anomalies at the feature level by sampling the learned quantized feature space, which allows a controlled generation of near-in-distribution anomalies. DSR achieves state-of-the-art results on the KSDD2 and MVTec anomaly detection datasets. The experiments on the challenging real-world KSDD2 dataset show that DSR significantly outperforms other unsupervised surface anomaly detection methods, improving the previous top-performing methods by 10% AP in anomaly detection and 35% AP in anomaly localization.