Research map / Vision-language and multimodal / Multimodal LLMs
EGVLR: Evidence-Grounded Vision–Language Reinforcement for Anomaly Reasoning
SHIH-CHIH LIN, Ying-Heng Lu, DONG-YOU YE, Shang-Hong Lai
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 PVE-FT visual pre-alignment (ablation) | Qwen3-VL-8B · MMAD | Loc. accuracy: 59.08 -> 71.90 | Improved |
| add KG-IT after PVE-FT (ablation) | EGVLR · MMAD | Avg accuracy: 77.84 -> 80.63 | Improved |
| add GS-DPO after KG-IT (ablation) | EGVLR · MMAD | FPR_img (%): 23.8 -> 13.9 | Improved |
| add GS-DPO after KG-IT (ablation) | EGVLR · MMAD | Avg accuracy: 80.63 -> 82.55 | Improved |
| use full decoupled reward instead of answer-only GRPO (ablation) | GS-DPO · MMAD | Loc. accuracy: 72.34 -> 76.85 | Improved |
| add box reward (ablation) | GS-DPO · MMAD | Loc. accuracy: 71.72 -> 76.85 | Improved |
| add BGE rationale reward (ablation) | GS-DPO · MMAD | FPR_img (%): 20.7 -> 13.9 | Improved |
| add semantic coupling reward (ablation) | GS-DPO · MMAD | FPR_img (%): 19.3 -> 13.9 | Improved |
| add 3x3 grid convention (ablation) | PVE-FT · MMAD | Loc. accuracy: 74.38 -> 76.85 | Improved |
| add local verification decoys (ablation) | PVE-FT · MMAD | FPR_img (%): 21.8 -> 13.9 | Improved |
| add null-hypothesis calibration (ablation) | PVE-FT · MMAD | FPR_img (%): 25.4 -> 13.9 | Improved |
| replace standard GRPO with decoupled GS-DPO reward (ablation) | GRPO · MMAD | FPR_img (%): 24.6 -> 13.9 | Improved |
Closest papers by meaning
- MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detection (ICLR 2025 · similarity 0.89)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.88)
- LogicAD: Explainable Anomaly Detection via VLM-based Text Feature Extraction (AAAI 2025 · similarity 0.88)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.87)
- Judo: A Juxtaposed Domain-oriented Multimodal Reasoner for Industrial Anomaly QA (ICLR 2026 · similarity 0.87)
- AD-FM: Multimodal LLMs for Anomaly Detection via Multi-Stage Reasoning and Fine-Grained Reward Optimization (AAAI 2026 · 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.
