Research map / Label-efficient and robust learning / OOD and anomaly detection
Fast Dynamic Prototypes for Unsupervised Anomaly Detection and Localization
Mingliang Li, Hanxi Li, Lin Yuanbo Wu, Changhong Liu, Xiaowei Zhao
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 |
|---|---|---|---|
| shrink prototype bank from 105 to 8192 (ablation) | Baseline · MVTec-AD | mAD: 90.9 -> 87.4 | Hurt |
| add Dynamic Prototype Adapter (ablation) | Baseline(8192) · MVTec-AD | mAD: 87.4 -> 91.3 | Improved |
| add Complementary Prototype Reconstruction (ablation) | Baseline + DPA · MVTec-AD | P-AP: 72.5 -> 73.0 | Improved |
| add segmentation head residual learning (ablation) | Baseline + DPA + CPR · MVTec-AD | P-AP: 73.0 -> 77.8 | Improved |
| add Hybrid Residual Learning (ablation) | Baseline + DPA + CPR + Seg. Head · MVTec-AD | mAD: 92.9 -> 93.5 | Improved |
| add auxiliary ResNet-18 complementary backbone (low CKA 0.3786 with DINOv3-S) (ablation) | FDP · MVTec-AD | CKA: 0.3786 vs 0.8767 and 0.9963 | Improved |
| dynamic extraction on small defects 0%-5% area (ablation) | FDP · MVTec-AD | I-AUROC: 98.1 -> 99.3 | Improved |
| dynamic extraction on large defects 20%-100% area (ablation) | FDP · MVTec-AD | P-AP: 77.2 -> 76.1 | Hurt |
| dynamic extraction on 5%-20% area (ablation) | FDP · MVTec-AD | I-AUROC: 100.0 -> 100.0 | No clear effect |
| replace backbone of INP-Former with DINOv3-small | INP-Former vs FDP · MVTec-AD | mAD: 90.8 -> 93.5 | Improved |
| use 512x512 instead of 392x392 input | FDP · MVTec-AD | mAD: 91.6 -> 93.5 | Improved |
| use 512x512 instead of 392x392 input | FDP · MVTec-AD | FPS: 358 -> 192 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Dinomaly — Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.87)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.86)
- Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection (CVPR 2025 · similarity 0.86)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.85)
- CostFilter-AD: Enhancing Anomaly Detection through Matching Cost Filtering (ICML 2025 · similarity 0.85)
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 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.
