Research map / Detection and segmentation / Open-set and few-shot detection
Towards Total Recall in Industrial Anomaly Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use locally aware patch features from intermediate levels 2 and 3 instead of relying on deep, abstract features | PatchCore · MVTec AD | image AUROC: 99.1 vs 97.9 | Helped |
| Subsample the memory bank with greedy coreset selection to 10% (ablation) | PatchCore · MVTec AD | inference time: 0.6 -> 0.22 | Helped |
| Reduce the memory bank to 1% with greedy coreset selection rather than using the full bank (ablation) | PatchCore · MVTec AD | inference time: 0.6 -> 0.17 | Helped |
| Reduce the memory bank to 1% with greedy coreset selection (ablation) | PatchCore · MVTec AD | image AUROC: 99.1 -> 99.0 | No clear effect |
| Reduce the memory bank to 1% with greedy coreset selection (ablation) | PatchCore · MVTec AD | pixel AUROC: 98.1 -> 98.0 | No clear effect |
| Use greedy coreset subsampling instead of random or learned-proxy subsampling (ablation) | PatchCore · MVTec AD | anomaly detection and localization performance | Helped |
| Increase patch-feature stride to 2 (ablation) | PatchCore · MVTec AD | image AUROC: 97.6% | Hurt |
| Increase patch-feature stride to 3 (ablation) | PatchCore · MVTec AD | image AUROC: 96.8% | Hurt |
| Add hierarchy 1 to hierarchy levels 2 and 3 (ablation) | PatchCore-1%, WideResNet101, image size 280 · MVTec AD | image AUROC: 99.4 -> 99.2 | Hurt |
| Add hierarchy 1 to hierarchy levels 2 and 3 (ablation) | PatchCore-1%, WideResNet101, image size 280 · MVTec AD | pixel AUROC: 98.2 -> 98.4 | Helped |
| Use an ensemble with higher-resolution images and larger backbones (ablation) | PatchCore-1% · MVTec AD | image AUROC: 99.0 -> 99.6 | Helped |
| Use the maximum patch distance without neighbor-based score reweighting (ablation) | PatchCore · MVTec AD | anomaly detection robustness | Hurt |
Closest papers by meaning
- InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection (NeurIPS 2026 · similarity 0.87)
- Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping (CVPR 2024 · similarity 0.86)
- Supervised Anomaly Detection for Complex Industrial Images (CVPR 2024 · similarity 0.86)
- MuSc: Zero-Shot Industrial Anomaly Classification and Segmentation with Mutual Scoring of the Unlabeled Images (ICLR 2024 · similarity 0.86)
- 3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly Detection (AAAI 2025 · similarity 0.86)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.85)
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.