Skip to content
Conference

Task-routed hierarchical patch memories for industrial anomaly detection and localization

Oct 2026 · International Conference on Image Processing, Object Detection and Tracking · Vol 14506, pp. 145061R - 145061R-7 · 0 citations · 12 references
Engineering

TL;DR

This work proposes TRHM-PatchCore, a normal only task-routed hierarchical patch-memory framework that applies fixed task routing rather than score fusion, and applies fixed task routing rather than score fusion.

Abstract

Industrial anomaly inspection must detect and localize defects, but patch-memory methods use one feature hierarchy for both outputs. This creates a task-dependent hierarchy conflict: deeper features aid image discrimination, whereas pixel localization requires spatial evidence weakened by resizing and aggregation. We propose TRHM-PatchCore, a normal only task-routed hierarchical patch-memory framework. At inference, one shared frozen pretrained WideResNet-50-2 extracts layers 1-3 once for two task-specific memory heads. Global aggregates layers 2 and 3, uses all extracted descriptors before 10% coreset selection, and supplies the image anomaly score. Detail aggregates layers 1 and 2, deterministically samples 2,048 descriptors per normal image before the same coreset procedure and supplies the pixel anomaly map. The final system applies fixed task routing rather than score fusion. On all eight MVTec AD 2 TESTpub categories, Global obtains macro-Image AUROC/AP of 0.7520/0.8350, whereas Detail raises macro AU-PRO@0.05 from 0.3681 to 0.4230 and AU-PRO@0.30 from 0.6096 to 0.6905. Across all eight categories, Detail retains macro-AU-PRO gains of +0.0477/+0.0714 under matched descriptor count, +0.0539/+0.0746 under matched storage, and +0.0540/+0.0769 under a matched nearest-neighbor search proxy at FPR limits 0.05/0.30, respectively. Across five matched-budget coreset seeds, the Detail-minus-Coarse AU-PRO gains are +0.0468 (95% CI 0.0437-0.0499) and +0.0692 (0.0667-0.0717) at FPR limits 0.05 and 0.30. On untouched Magnetic Tile, it matches standard PatchCore image metrics; AU-PRO@0.05/@0.30 is 0.5724/0.8124 versus 0.5604/0.7960.

View source

Similar papers

Conference Open access Sep 2026

Recovering Total Recall: Multi-Scale Feature Fusion for High-Resolution Industrial Anomaly Localization

PatchCore, a memory-bank-based anomaly detector, is near-saturated on the original MVTec AD benchmark but deteriorates sharply on the recently released MVTec AD 2, where its mean Area-Under-the-Per-Region-Overlap at 5% FPR (AU-PRO@5%) drops to single digits in our reproduction. We investigate this regression on a high-...

Ya-Yang Qian · 0 citations
Open access Sep 2026

Relation Prototype Re-Scoring for CLIP-Based Logical Anomaly Detection and Localization

CLIP-based anomaly detectors have markedly advanced training-free and zero-shot industrial anomaly detection and localization, yet their predictions remain dominated by patch-wise vision–language similarity or anomaly-aware feature scoring. This formulation is intrinsically limited for logical anomalies, in which every...

Hanhoon Park · 0 citations
Conference Open access 2026

Lightweight Augmentation of SimpleNet for Industrial Anomaly Detection via Residual Spatial Attention

Industrial anomaly detection requires both image-level recognition and pixel-level localization when anomaly samples are scarce. SimpleNet has become a representative lightweight baseline because of its single-stream inference pipeline, compact structure and low deployment cost. However, its multi-layer features lack a...

Hao Ni · 0 citations
Open access Sep 2026

A Comparative Study of Local-Global Anomaly-Map Fusion Strategies in EfficientAD

Industrial anomaly detection must identify both local structural defects and global logical violations while preserving localization quality and practical inference cost. EfficientAD provides complementary local student-teacher and global autoencoder-student anomaly maps, yet its fixed equal average assumes that both m...

Qin-Yu Xiao · 0 citations
Open access Nov 2026

Unsupervised Industrial Anomaly Detection and Localization via Anomaly Score Fusion of Dual-Layer DINOv3 Features

Unsupervised industrial anomaly detection uses normal training images to identify and localize defects. The challenge addressed here is localizing small defects amid normal appearance variation without requiring different feature layers to share one normal reference. We propose dual-layer DINOv3 anomaly-score fusion (D...

Bi-Han Liang, Wei Liang, Hai-Hua Liang et al. · 0 citations
Open access 2026

Few-Shot Anomaly Detection on MVTec AD 2: A Systematic Study and a Strong Frozen Baseline

This study identifies three empirical findings: two deep layers improve localisation over the common four-layer concatenation; pooling the highest-scoring patches outperforms averaging all patches; and the tested greedy coreset reference strategy degrades pixel-level performance at both evaluated reference budgets.

Wei-Lin Tu · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.