How Low Can You Go? Surfacing Prototypical In-Distribution Samples for Unsupervised Anomaly Detection
This work addresses the challenge of reducing labeling efforts in anomaly detection for domains like computer vision, industrial defect detection, and medicine, but it is incremental as it builds on existing UAD methods.
The paper tackles the problem of unsupervised anomaly detection (UAD) by showing that using extremely few training samples can match or surpass performance with full datasets, and proposes an unsupervised method to identify prototypical samples to boost UAD performance, achieving better results than full training in 25 out of 67 categories with just 25 selected samples.
Unsupervised anomaly detection (UAD) alleviates large labeling efforts by training exclusively on unlabeled in-distribution data and detecting outliers as anomalies. Generally, the assumption prevails that large training datasets allow the training of higher-performing UAD models. However, in this work, we show that UAD with extremely few training samples can already match -- and in some cases even surpass -- the performance of training with the whole training dataset. Building upon this finding, we propose an unsupervised method to reliably identify prototypical samples to further boost UAD performance. We demonstrate the utility of our method on seven different established UAD benchmarks from computer vision, industrial defect detection, and medicine. With just 25 selected samples, we even exceed the performance of full training in $25/67$ categories in these benchmarks. Additionally, we show that the prototypical in-distribution samples identified by our proposed method generalize well across models and datasets and that observing their sample selection criteria allows for a successful manual selection of small subsets of high-performing samples. Our code is available at https://anonymous.4open.science/r/uad_prototypical_samples/