Research map / Vision-language and multimodal / CLIP and zero-shot
Robust Zero-shot Anomaly Detection under Limited Auxiliary Anomaly Priors
Guanyu Lu, Fang Zhou, Cheqing Jin
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 |
|---|---|---|---|
| pretrain on DTD (texture-only anomalies) instead of MVTec | baseline ZSAD models · 12-dataset average | AP: drops of 9.9%-23.3% AP | Hurt |
| use DIVE instead of baselines under DTD auxiliary data | DIVE · 12-dataset classification avg (DTD-pretrained) | AUROC: 81.6 (TPS) -> 87.3 | Improved |
| add DIVE under sufficient anomaly diversity | DIVE · 12-dataset classification avg (MVTec-pretrained) | AUROC: 89.1 (AdaCLIP) -> 90.1 | Improved |
| add text embedding injection (shallow + deep) (ablation) | DIVE · MVTec classification (DTD-pretrained) | AP: 69.5 -> 73.9 | Improved |
| add deep-level injection of LLM descriptions (ablation) | DIVE · Visa classification (DTD-pretrained) | AP: 46.0 -> 50.0 | Improved |
| add shallow-level text injection (ablation) | DIVE · Visa classification (DTD-pretrained) | AP: 49.1 -> 50.0 | Improved |
| add disentanglement mechanism (ablation) | DIVE · Visa classification (DTD-pretrained) | AP: 48.4 -> 50.0 | Improved |
| add disentanglement mechanism (ablation) | DIVE · MVTec segmentation (DTD-pretrained) | AUROC: 89.9 -> 90.5 | Improved |
| DIVE vs AA-CLIP on ISIC under RCPRO | DIVE · ISIC segmentation (DTD-pretrained) | RCPRO: 89.3 (AA-CLIP) vs 74.2 (DIVE) | Hurt |
| vary prompting depth P over 3, 6, 9, 12 (ablation) | DIVE · MVTec and Visa | AUROC/AP | Mixed |
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.
- AA-CLIP — AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025) · better on 3 of 3 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
- VarProtoAD: Variational Prototype-Conditioned Prompting for Zero-Shot Anomaly Detection (ECCV 2026 · similarity 0.90)
- VisualAD: Language-Free Zero-Shot Anomaly Detection via Vision Transformer (CVPR 2026 · similarity 0.90)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detection (CVPR 2025 · similarity 0.87)
- Unseen Visual Anomaly Generation (CVPR 2025 · similarity 0.87)
- AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection (ICLR 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.
