ProCon: Projection-Consistency Memory for Training-Free Anomaly Detection

2026-07-06

AI Summary

This work proposes a training-free anomaly detection framework that overcomes the limitations of existing memory-based methods, which rely on hard nearest-neighbor retrieval and are thus susceptible to erroneous matches that hinder reliable validation of whether test patches receive consistent support from locally normal neighbors. The proposed approach reformulates memory retrieval as decoder-free reconstruction by softly projecting test image patches into the normal memory vector space and using the resulting projection residuals as the anomaly criterion. Key innovations include replacing hard retrieval with soft projection, hierarchical memory construction via seed perturbation, median aggregation of residuals, and deep consensus fusion of multi-layer residual maps. The method achieves state-of-the-art image-level AUROC scores of 99.8%, 99.2%, and 93.2% on MVTec-AD, VisA, and Real-IAD, respectively, demonstrating superior performance across seven standard evaluation metrics.

Abstract

Memory-based anomaly detection is attractive because it localizes defects from normal images without training a decoder or synthesizing pseudo anomalies. However, most memory methods still use the memory bank as a nearest-neighbor lookup table: a test patch is treated as normal if it has one nearby normal anchor. This hard retrieval view is vulnerable to false-normal matches and does not test whether the patch is consistently supported by a local normal neighborhood. We propose ProCon, a training-free framework that turns memory retrieval into decoder-free reconstruction. ProCon softly projects each test patch onto nearby normal memory vectors and uses the projection residual as anomaly evidence. To stabilize this residual, it constructs seed-perturbed layer-wise memories, aggregates bank residuals by a median, and fuses depth-specific residual maps by layer consensus. ProCon requires no decoder training, backbone fine-tuning, learned fusion weights, or pseudo-anomaly supervision. Across MVTec-AD, VisA, and Real-IAD under the single-category evaluation protocol, ProCon achieves strong image- and pixel-level performance under seven standard metrics, including image AUROC scores of 99.8%, 99.2%, and 93.2%, respectively. Ablations show that the gains come from replacing hard retrieval with soft normal projection and stabilizing the residuals through memory and depth consensus. The code is available at https://github.com/jw-chae/Procon

Problem

• anomaly detection
• memory-based
• training-free
• projection consistency
• normal neighborhood

Innovation

• projection-consistency
• training-free
• memory-based anomaly detection
• soft projection
• layer consensus