Research map / Label-efficient and robust learning / OOD and anomaly detection
RAID: Retrieval-Augmented Anomaly Detection
Mingxiu Cai, Zhe Zhang, Gaochang Wu, Tianyou Chai, Xiatian Zhu
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 the full guided MoE filter to retrieval-only matching (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 97.9 / 97.5 -> 99.4 / 98.6 | Improved |
| add first-stage MoE_g guidance fusion (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 98.5 / 97.6 -> 99.2 / 98.4 | Improved |
| add Router_C sparse routing to the second-stage MoE (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 98.0 / 97.5 -> 99.4 / 98.6 | Improved |
| use hierarchical instead of flat retrieval (ablation) | RAID · MVTec-AD | inference time (s): 0.267 -> 0.062 | Improved |
| use hierarchical instead of flat retrieval (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 99.4 / 98.7 -> 99.4 / 98.6 | No clear effect |
| add instance token level to the hierarchy (ablation) | RAID · VisA | I-AUROC / P-AP: 79.5 / 30.1 -> 94.9 / 45.2 | Improved |
| add semantic prototype level to the hierarchy (ablation) | RAID · VisA | I-AUROC / P-AP: 91.9 / 40.0 -> 94.9 / 45.2 | Improved |
| add class prototype level to the hierarchy (ablation) | RAID · VisA | I-AUROC / P-AP: 93.3 / 43.2 -> 94.9 / 45.2 | Improved |
| increase template count from 20 to 80 (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 98.1 / 97.7 -> 99.4 / 98.6 | Improved |
| use all templates instead of 80 (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 99.4 / 98.6 -> 99.3 / 98.6 | No clear effect |
| increase guidance experts E_g from 2 to 3 (with E_C = 3) (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 99.0 / 98.1 -> 99.4 / 98.6 | Improved |
| increase guidance experts E_g from 3 to 4 (with E_C = 3) (ablation) | RAID · MVTec-AD | I-AUROC / P-AUROC: 99.4 / 98.6 -> 98.8 / 97.9 | Hurt |
Closest papers by meaning
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.88)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.88)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.87)
- CostFilter-AD: Enhancing Anomaly Detection through Matching Cost Filtering (ICML 2025 · similarity 0.86)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.86)
- Unseen Visual Anomaly Generation (CVPR 2025 · 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.
