What Memory Composition Does Not Tell Us About Anomaly Detection
2026-08-24 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors studied how memory-based anomaly detectors pick reference patches from training data to spot unusual patterns. They found that some methods tend to include rare, contaminated patches, which can mislead the detector. They introduced CLEANCON, a new technique that reduces such contamination in memory while keeping other factors constant. Although CLEANCON lowered contamination and improved overall detection scores, the authors noticed that having less contamination did not always mean better performance. This suggests that simply reducing contamination isn't enough to predict how well the detector will work.
anomaly detectionmemory-based detectorstraining contaminationpatch selectioncoverage selectorsCLEANCONrepresentation learningprecision-average precision (P-AP)sparse contaminationcross-image support
Authors
Joongwon Chae, Runming Wang, Peiwu Qin
Abstract
Memory-based anomaly detectors store nominal training patches and score test patches against this memory. A patch selected for coverage therefore becomes a nor- mal reference without a separate check that geometric rarity makes it safe to trust. We probe this coupling with sparse training contamination. Under fixed representa- tions and memory budgets, we compare random, medoid, local, and global coverage selectors. We then use CLEANCON, an out-of-bag cross-image support gate that changes candidate-image eligibility while fixing the representation, absolute mem- ory size, builder, and inference rule. Global coverage strongly over-represents sparse contamination. CLEANCON reduces final-memory contamination to approx- imately zero and increases category-macro P-AP in all 12 matched comparisons. Yet along a retention sweep, the lowest-contamination memory does not attain the highest P-AP; performance continues to improve while contamination rises. Mem- ory contamination therefore does not order the resulting memories by P-AP