Research map / Detection and segmentation / Open-set and few-shot detection
Reasoning-Driven Anomaly Detection and Localization with Image-Level Supervision
Yizhou Jin, Yuezhu Feng, Jinjin Zhang, Peng Wang, Qingjie Liu, Yunhong Wang
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 consistency reward (CGRO) on top of R1 rewards (ablation) | Qwen2.5-VL-7B + ReAL · Zero-shot IAD image-level avg | AUROC: 80.0 -> 83.9 | Improved |
| add consistency reward (CGRO) on top of R1 rewards (ablation) | Qwen2.5-VL-3B + ReAL · Zero-shot IAD image-level avg | AUROC: 58.1 -> 67.6 | Improved |
| add consistency reward on top of R1 rewards (ablation) | Qwen2.5-VL-7B + ReAL · Zero-shot IAD pixel-level avg | AUROC: 78.5 -> 80.7 | Improved |
| add consistency reward on top of R1 rewards (ablation) | Qwen2.5-VL-3B + ReAL · MVTec pixel-level | AUROC: 63.2 -> 62.1 | Hurt |
| add ReAL token selection to CGRO model (ablation) | Qwen2.5-VL-7B · pixel-level macro avg | AUROC: 72.7 -> 80.7 | Improved |
| apply ReAL to untrained model without CGRO (ablation) | Qwen2.5-VL-7B · pixel-level macro avg | AUROC: 64.7 -> 61.7 | Hurt |
| add semantic relevance filter ST to spatial entropy filter (ablation) | ReAL · pixel-level macro avg | AUROC: 74.1 -> 80.7 | Improved |
| add spatial entropy filter to semantic relevance filter (ablation) | ReAL · pixel-level macro avg | AUROC: 76.7 -> 80.7 | Improved |
| add RL fine-tuning with image-level labels (R1) | Qwen2.5-VL-7B · Zero-shot IAD image-level avg | AUROC: 63.4 -> 80.0 | Improved |
Closest papers by meaning
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · similarity 0.89)
- AD-FM: Multimodal LLMs for Anomaly Detection via Multi-Stage Reasoning and Fine-Grained Reward Optimization (AAAI 2026 · similarity 0.88)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- Vad-R1: Towards Video Anomaly Reasoning via Perception-to-Cognition Chain-of-Thought (NeurIPS 2025 · similarity 0.87)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.87)
- PANDA: Towards Generalist Video Anomaly Detection via Agentic AI Engineer (NeurIPS 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.
