Research map / Label-efficient and robust learning / OOD and anomaly detection
MMR-AD: A Large-Scale Multimodal Dataset for Benchmarking General Anomaly Detection with Multimodal Large Language Models
Xincheng Yao, Zefeng Qian, Chao Shi, Jiayang Song, Chongyang 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 reasoning text (CoT) to training data (ablation) | Anomaly-R1-7B · MVTecAD | detection accuracy: 81.3 -> 88.8 | Improved |
| add reasoning text (CoT) to training data (ablation) | Anomaly-R1-7B · VisA | detection accuracy: 66.1 -> 74.9 | Improved |
| add normal reference image to the input (ablation) | Anomaly-R1-7B · MVTecAD | localization accuracy (IoU>=0.1): 62.1 -> 66.3 | Improved |
| add normal reference image to the input (ablation) | Anomaly-R1-7B · VisA | detection accuracy: 62.5 -> 74.9 | Improved |
| add CoT cold-start SFT before RL (ablation) | Anomaly-R1-7B · MVTecAD | localization accuracy (IoU>=0.1): 12.3 -> 66.3 | Improved |
| add CoT cold-start SFT before RL (ablation) | Anomaly-R1-7B · VisA | localization accuracy (IoU>=0.1): 5.3 -> 36.4 | Improved |
| add reinforcement learning (GRPO) after SFT (ablation) | Anomaly-R1-7B · MVTecAD | detection accuracy: 84.5 -> 88.8 | Improved |
| add reinforcement learning (GRPO) after SFT (ablation) | Anomaly-R1-7B · VisA | detection accuracy: 73.3 -> 74.9 | Improved |
| add contrastive sampling to GRPO (ablation) | Anomaly-R1-7B · MVTecAD | detection accuracy: 87.7 -> 88.8 | Improved |
| add contrastive sampling to GRPO (ablation) | Anomaly-R1-7B · VisA | detection accuracy: 73.7 -> 74.9 | Improved |
| add anomaly-related domain knowledge to the prompt (ablation) | Anomaly-R1-7B · MVTecAD | detection accuracy: 88.8 -> 91.0 | Improved |
| add anomaly-related domain knowledge to the prompt (ablation) | Anomaly-R1-7B · VisA | detection accuracy: 74.9 -> 79.0 | Improved |
Closest papers by meaning
- AD-FM: Multimodal LLMs for Anomaly Detection via Multi-Stage Reasoning and Fine-Grained Reward Optimization (AAAI 2026 · similarity 0.90)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.90)
- MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detection (ICLR 2025 · similarity 0.89)
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024 · similarity 0.88)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.88)
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · 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.
