Locally varying distance transform for unsupervised visual anomaly detection

ECCV 2022  ·  Wen-Yan Lin, Zhonghang Liu, Siying Liu ·

Unsupervised anomaly detection on image data is notoriously unstable. We believe this is because many classical anomaly detectors implicitly assume data is low dimensional. However, image data is always high dimensional. Images can be projected to a low dimensional embedding but such projections rely on global transformations that truncate minor variations. As anomalies are rare, the final embedding often lacks the key variations needed to distinguish anomalies from normal instances. This paper proposes a new embedding using a set of locally varying data projections, with each projection responsible for persevering the variations that distinguish a local cluster of instances from all other instances. The locally varying embedding ensures the variations that distinguish anomalies are preserved, while simultaneously allowing the probability that an instance belongs to a cluster, to be statistically inferred from the one-dimensional, local projection associated with the cluster. Statistical agglomeration of an instance’s cluster membership probabilities, creates a global measure of its affinity to the dataset and causes anomalies to emerge, as instances whose affinity scores are surprisingly low.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Unsupervised Anomaly Detection with Specified Settings -- 30% anomaly ASSIRA Cat Vs Dog LVAD AUC-ROC 0.780 # 3
Unsupervised Anomaly Detection with Specified Settings -- 1% anomaly Cats and Dogs LVAD AUC-ROC 0.978 # 2
Unsupervised Anomaly Detection with Specified Settings -- 10% anomaly Cats and Dogs LVAD AUC-ROC 0.927 # 4
Unsupervised Anomaly Detection with Specified Settings -- 0.1% anomaly Cats and Dogs LVAD AUC-ROC 0.981 # 2
Unsupervised Anomaly Detection with Specified Settings -- 20% anomaly Cats and Dogs LVAD AUC-ROC 0.851 # 4
Unsupervised Anomaly Detection with Specified Settings -- 20% anomaly cifar10 LVAD AUC-ROC 0.854 # 3
Unsupervised Anomaly Detection with Specified Settings -- 10% anomaly CIFAR-10 LVAD AUC-ROC 0.903 # 1
Unsupervised Anomaly Detection with Specified Settings -- 30% anomaly CIFAR-10 LVAD AUC-ROC 0.816 # 3
Unsupervised Anomaly Detection with Specified Settings -- 0.1% anomaly CIFAR-10 LVAD AUC-ROC 0.930 # 1
Unsupervised Anomaly Detection with Specified Settings -- 1% anomaly CIFAR-10 LVAD AUC-ROC 0.940 # 1
Unsupervised Anomaly Detection with Specified Settings -- 10% anomaly Fashion-MNIST LVAD AUC-ROC 0.899 # 2
Unsupervised Anomaly Detection with Specified Settings -- 30% anomaly Fashion-MNIST LVAD AUC-ROC 0.868 # 2
Unsupervised Anomaly Detection with Specified Settings -- 1% anomaly Fashion-MNIST LVAD AUC-ROC 0.909 # 2
Unsupervised Anomaly Detection with Specified Settings -- 20% anomaly Fashion-MNIST LVAD AUC-ROC 0.884 # 2
Unsupervised Anomaly Detection with Specified Settings -- 0.1% anomaly Fashion-MNIST LVAD AUC-ROC 0.896 # 3
Unsupervised Anomaly Detection with Specified Settings -- 20% anomaly MNIST LVAD AUC-ROC 0.923 # 1
Unsupervised Anomaly Detection MNIST LVAD AUROC 0.937 # 1
Unsupervised Anomaly Detection with Specified Settings -- 0.1% anomaly MNIST LVAD AUC-ROC 0.974 # 1
Unsupervised Anomaly Detection with Specified Settings -- 1% anomaly MNIST LVAD AUC-ROC 0.948 # 1
Unsupervised Anomaly Detection with Specified Settings -- 10% anomaly MNIST LVAD AUC-ROC 0.938 # 1
Unsupervised Anomaly Detection with Specified Settings -- 30% anomaly MNIST LVAD AUC-ROC 0.904 # 1
Unsupervised Anomaly Detection STL-10 LVAD AUC-ROC 0.996 # 1
Unsupervised Anomaly Detection with Specified Settings -- 1% anomaly STL-10 LVAD AUC-ROC 0.993 # 1
Unsupervised Anomaly Detection with Specified Settings -- 30% anomaly STL-10 LVAD AUC-ROC 0.977 # 2
Unsupervised Anomaly Detection with Specified Settings -- 20% anomaly STL-10 LVAD AUC-ROC 0.983 # 2
Unsupervised Anomaly Detection with Specified Settings -- 0.1% anomaly STL-10 LVAD AUC-ROC 0.998 # 1
Unsupervised Anomaly Detection with Specified Settings -- 10% anomaly STL-10 LVAD AUC-ROC 0.979 # 2

Methods


No methods listed for this paper. Add relevant methods here