Research map / Label-efficient and robust learning / OOD and anomaly detection
MRAD: Zero-Shot Anomaly Detection with Memory-Driven Retrieval
Chaoran Xu, Chengkan Lv, Qiyu Chen, Feng Zhang, Zhengtao Zhang
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 lightweight two-linear-layer retrieval fine-tuning (ablation) | MRAD-TF · 16-dataset average | I-AUROC: 81.0 -> 92.0 | Improved |
| add lightweight two-linear-layer retrieval fine-tuning (ablation) | MRAD-TF · 16-dataset average | P-AUROC: 85.5 -> 91.9 | Improved |
| add prior-guided dynamic prompt biases (MRAD-CLIP) (ablation) | MRAD-FT · 16-dataset average | PRO: 78.3 -> 80.1 | Improved |
| add prior-guided dynamic prompt biases (MRAD-CLIP) (ablation) | MRAD-FT · 16-dataset average | I-AUROC: 92.0 -> 92.7 | Improved |
| bias prompts with CLIP class token instead of static prompts (ablation) | MRAD-CLIP · MVTec-AD | P-AUROC: 91.5 -> 86.2 | Hurt |
| add cross-patch attention bias to static prompts (ablation) | MRAD-CLIP · MVTec-AD | P-AUROC: 91.5 -> 92.5 | Improved |
| add dual normal+anomalous prior bias instead of anomaly prior only (ablation) | MRAD-CLIP · Average over datasets pixel-level | P-AUROC: 92.0 -> 92.7 | Improved |
| add dual prior bias instead of cross-patch bias (ablation) | MRAD-CLIP · Average over datasets image-level | I-AUROC: 92.1 -> 92.7 | Improved |
| use BTAD instead of VisA as auxiliary dataset (ablation) | MRAD-CLIP · MVTec-AD | P-AUROC: 93.0 -> 91.0 | Hurt |
| use SDD instead of VisA as auxiliary dataset (ablation) | MRAD-CLIP · MVTec-AD | I-AUROC: 94.0 -> 87.8 | Hurt |
| add image-level memory to pixel-level-only memory (ablation) | MRAD · image-level detection | I-AUROC | Improved |
| add pixel-level memory to image-level-only memory (ablation) | MRAD · image-level detection | I-AUROC | Improved |
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 0.89)
- DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 2025 · similarity 0.88)
- AA-CLIP: Enhancing Zero-Shot Anomaly Detection via Anomaly-Aware CLIP (CVPR 2025 · similarity 0.87)
- Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detection (CVPR 2025 · similarity 0.87)
- PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection (CVPR 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.
