Research map / Label-efficient and robust learning / OOD and anomaly detection
Hard-Normal Example-Aware Template Mutual Matching for Industrial Anomaly Detection
Zixuan Chen, Xiaohua Xie, Lingxiao Yang, Jianhuang Lai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add backward matching to forward ATM (bidirectional ATMM) (ablation) | F-ATM · MVTec AD (ALL templates) | image AUROC: 99.2 -> 99.8 | Improved |
| add forward matching to backward ATM (ablation) | B-ATM · MVTec AD (ALL templates) | image AUROC: 99.0 -> 99.8 | Improved |
| use affine-invariant template matching instead of patch-level matching (ablation) | Patch-level matching · MVTec AD (ALL templates) | PRO: 94.9 -> 96.4 | Improved |
| use affine-invariant template matching instead of pixel-level matching (ablation) | Pixel-level matching · MVTec AD (ALL templates) | image AUROC: 96.6 -> 99.8 | Improved |
| combine forward and backward ATM (ablation) | F-ATM · MVTec LOCO | average image AUROC: 86.1 -> 88.1 | Improved |
| combine forward and backward ATM (logical anomalies) (ablation) | F-ATM · MVTec LOCO | logical image AUROC: 78.8 -> 83.2 | Improved |
| combine forward and backward ATM (structural anomalies) (ablation) | F-ATM · MVTec LOCO | structural image AUROC: 93.4 -> 92.9 | Hurt |
| use hard-normal-aware PTS_O instead of pixel cluster centres PC_O (ablation) | ATMM · MVTec AD (60 sheets) | image AUROC: 98.9 -> 99.5 | Improved |
| use PTS_K instead of pixel random selection (ablation) | ATMM · MVTec AD (60 sheets) | image AUROC: 98.4 -> 99.3 | Improved |
| use pixel-level instead of image-level selection (PC_K vs IC_K) (ablation) | ATMM · MVTec AD (60 sheets) | image AUROC: 97.1 -> 98.7 | Improved |
| compress template set to 60 sheets with PTS | HETMM · MVTec AD (ALL templates) | image AUROC: 99.8 -> 99.5 | Hurt |
| compress template set to 60 sheets with PTS | HETMM · MVTec AD (ALL templates) | inference speed (FPS): 4.7 FPS -> 26.1 FPS | Improved |
Closest papers by meaning
- Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection (CVPR 2025 · similarity 0.88)
- DeMuS: Learning Decoupled Matching and Scoring for Batch Zero-Shot Industrial Anomaly Detection (ECCV 2026 · similarity 0.88)
- MATCH: Flow Matching for Multi-View Anomaly Detection (ECCV 2026 · similarity 0.88)
- InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection (NeurIPS 2026 · similarity 0.87)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.86)
- MuSc: Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
