Research map / Label-efficient and robust learning / OOD and anomaly detection
Bidirectional Multimodal Prompt Learning with Scale-Aware Training for Few-Shot Multi-Class Anomaly Detection
Yujin Lee, Sewon Kim, Daeun Moon, Seoyoon Jang, Hyunsoo Yoon
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 bidirectional text-image prompt coupling (ablation) | AnoPLe · MVTec-AD / VisA / Real-IAD 1-shot | image AUROC: 93.0/90.3 -> 94.5 on MVTec-AD | Improved |
| add one-way I->T coupling instead of bidirectional (ablation) | AnoPLe · MVTec-AD 1-shot | image AUROC: 94.5 -> 91.2 | Hurt |
| add alignment loss between local and global features (ablation) | AnoPLe · Real-IAD 1-shot | image AUROC: 73.3 -> 81.2 | Improved |
| use weighted-sum alignment instead of mean alignment (ablation) | AnoPLe · Real-IAD 1-shot | image AUROC: 74.2 -> 81.2 | Improved |
| add multi-scale training (ablation) | AnoPLe · MVTec-AD 1-shot | image AUROC: 89.9 -> 91.8 | Improved |
| add scale-aware prefix on top of multi-scale training (ablation) | AnoPLe · MVTec-AD 1-shot | image AUROC: 91.8 -> 94.5 | Improved |
| use class name instead of generic [object] token (ablation) | AnoPLe · Real-IAD 1-shot | image AUROC: 76.0 -> 81.2 | Improved |
| remove anomaly descriptions from PromptAD prompt pool (ablation) | PromptAD · MVTec-AD screw | image AUROC: 55.7 -> 44.7 | Hurt |
| hold out one class during training (AnoPLe vs INP-Former) | AnoPLe · MVTec-AD held-out | image AUROC: INP-Former 67.7 vs AnoPLe 87.5 | Improved |
| AnoPLe vs vision-only INP-Former on all-seen 4-shot | AnoPLe · Real-IAD 4-shot | image AUROC: 88.3 vs 83.2 | Hurt |
| apply AnoPLe to medical anomaly detection without adaptation | AnoPLe · BMAD Brain MRI 4-shot | image AUROC: UniVAD 82.6 vs AnoPLe 79.8 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- UniVAD — UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025) · better on 0 of 1 shared result
- PromptAD — PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection (CVPR 2024) · better on 1 of 1 shared result
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
- PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection (CVPR 2024 · similarity 0.90)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.89)
- Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detection (CVPR 2025 · similarity 0.88)
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024 · similarity 0.88)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.87)
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.
