Research map / Vision-language and multimodal / CLIP and zero-shot
Fine-grained Abnormality Prompt Learning for Zero-shot Anomaly Detection
Jiawen Zhu, Yew-Soon Ong, Chunhua Shen, Guansong Pang
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 CAP (compound abnormality prompts) to AnomalyCLIP (ablation) | AnomalyCLIP · Industrial avg image-level | AUROC: 85.0 -> 88.1 | Improved |
| add orthogonal constraint loss to CAP (ablation) | CAP · Industrial avg image-level | AUROC: 87.2 -> 88.1 | Improved |
| add DAP (data-dependent abnormality prior) to AnomalyCLIP (ablation) | AnomalyCLIP · Medical avg image-level | AUROC: 87.7 -> 90.2 | Improved |
| add abnormality prior selection loss to DAP (ablation) | DAP · Industrial avg image-level | AUROC: 86.5 -> 86.9 | Improved |
| combine CAP and DAP (ablation) | AnomalyCLIP · Industrial avg image-level | AUROC: 88.1 -> 88.5 | Improved |
| use abnormal patch token prior (P) instead of image token prior (I) (ablation) | FAPrompt · Industrial avg image-level | AUROC: 87.3 -> 88.5 | Improved |
| replace CoCoOp image-token prior with FAPrompt components (Lprior, CAP) (ablation) | CoCoOp · Industrial avg image-level | AUROC: 73.1 -> 87.3 | Improved |
| use CAP prompts instead of ensembling AnomalyCLIP with different seeds (ablation) | AnomalyCLIP Ensemble · Industrial avg image-level | AUROC: 85.5 -> 88.5 | Improved |
| average abnormality prompts instead of best-matched prompt selection (ablation) | FAPrompt · Industrial avg image-level | AUROC: 85.1 -> 88.5 | Improved |
| average prompts instead of learned weighted aggregation (ablation) | FAPrompt · Industrial avg image-level | AUROC: 87.7 -> 88.5 | Improved |
| increase number of abnormality prompts K beyond 10 (ablation) | FAPrompt · Industrial datasets | AUROC | Mixed |
| increase number of selected patch tokens M beyond 10 (ablation) | FAPrompt · Industrial datasets | AUROC | Mixed |
Closest papers by meaning
- AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection (ICLR 2024 · similarity 0.89)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- Aligning and Prompting Anything for Zero-Shot Generalized Anomaly Detection (AAAI 2025 · similarity 0.87)
- Beyond Normal References: Discriminative Few-Shot Anomaly Detection (NeurIPS 2026 · similarity 0.87)
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024 · similarity 0.87)
- DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 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.
