ShiftSplit-AD: Separating Domain Shift from Defects in Foundation-Feature Visual Anomaly Detection
Computer Vision and Pattern Recognition
Summary
The authors studied a way to improve visual anomaly detection by separating normal and abnormal features in images. Their method, called ShiftSplit-AD, breaks down feature differences into common patterns and rare events to better spot defects despite changes in how images are taken. They found this helps detect anomalies better on some tests but can also miss defects when the defect patterns look like normal changes. So, while their approach reduces false alarms from image shifts, it still struggles to keep all defect details in tricky cases.
Visual anomaly detectionFoundation modelsFeature residualsDomain shiftDINOv2Low-rank decompositionSparse componentsAUROCAUPRCMVTec dataset
Authors
Muhamathu Ameer Ali Aacaas Muhamath
Abstract
Visual anomaly detectors based on frozen foundation-model features commonly score distances from test patches to a memory of normal features. Benign acquisition changes can also enlarge these distances, confounding domain variation with defects. We investigate whether structured decomposition of nearest-normal DINOv2 residuals can suppress shift-induced evidence while retaining unseen defects. ShiftSplit-AD decomposes the patch residual matrix into low-rank and row-sparse components and scores the sparse component, with an optional low-rank/sparse fusion. The experiments expose a central trade-off rather than a universal separation: genuine defects can contain correlated, low-dimensional structure, so filtering broad residual activity may also remove defect information. On AeBAD-S, using settings fixed after Bottle development, sparse-only scoring improves image AUROC from 0.6780 to 0.7294 and AUPRC from 0.8052 to 0.8465. Paired bootstrap 95% intervals for the improvements are [0.0238, 0.0808] and [0.0170, 0.0650], respectively. However, sparse-only scoring reduces mean clean AUROC from 0.9890 to 0.9133 on four held-out MVTec categories and degrades Bottle localization. These findings show that residual decomposition can help when domain shift strongly contaminates anomaly evidence, but preserving defect structure remains the limiting problem.