Search papers, labs, and topics across Lattice.
This study investigates the impact of memory contamination on the performance of memory-based anomaly detectors, specifically examining how different coverage selectors influence the representation of nominal training patches. By introducing CLEANCON, an out-of-bag cross-image support gate, the authors effectively reduce memory contamination to nearly zero while enhancing category-macro P-AP across various comparisons. The findings reveal a counterintuitive relationship where lower contamination does not necessarily correlate with higher performance, indicating that memory contamination does not dictate the effectiveness of anomaly detection models.
CLEANCON achieves near-zero memory contamination while paradoxically showing that less contamination doesn't guarantee better anomaly detection performance.
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.Code is publicly available at https://github.com/jw-chae/cleancon.
