Research map / Detection and segmentation / Open-set and few-shot detection
AG-VAS: Anchor-Guided Zero-Shot Visual Anomaly Segmentation with Large Multimodal Models
Zhen Qu, Xian Tao, Xiaoyi Bao, Dingrong Wang, ShiChen Qu, Zhengtao Zhang, Xingang 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 [SEG] absolute anchor (ablation) | AG-VAS · MVTec-AD | pixel AP: 49.1 -> 51.0 | Improved |
| add [NOR][ANO] relative anchors (ablation) | AG-VAS · MVTec-AD | IoU_nor: 52.1 -> 87.7 | Improved |
| add SPAM (ablation) | AG-VAS · MVTec-AD | IoU_nor: 70.6 -> 87.7 | Improved |
| add SPAM (ablation) | AG-VAS · MVTec-AD | pixel AP: 46.5 -> 51.0 | Improved |
| train with Anomaly-Instruct20K (ablation) | AG-VAS · MVTec-AD | IoU_ano: 39.9 -> 44.8 | Improved |
| add Anomaly-Seg20K auxiliary images (ablation) | AG-VAS · MVTec-AD | IoU_nor: 53.5 -> 87.7 | Improved |
| add general segmentation data to joint training (ablation) | AG-VAS · MVTec-AD | pixel AP: 36.1 -> 51.0 | Improved |
| describe anomaly before segmenting (Describe-then-segment) vs direct (ablation) | AG-VAS · MVTec-AD | IoU_nor: 87.7 -> 58.8 | Hurt |
| prepend normal-appearance prior (Describe-then-segment-plus) vs direct (ablation) | AG-VAS · MVTec-AD | pixel AP: 51.0 -> 51.9 | Improved |
| segment-then-explain vs direct (ablation) | AG-VAS · MVTec-AD | pixel AP: 51.0 -> 48.6 | Hurt |
| anchor-guided LMM vs retrained LISA with same backbone | LLaVA-OneVision-7B · ColonDB | pixel AP: 67.2 -> 70.7 | Improved |
| anchor-guided LMM vs retrained LISA | LLaVA-OneVision-7B · RSDD | IoU_ano: 34.6 -> 36.9 | Improved |
Closest papers by meaning
- MetaUAS: Universal Anomaly Segmentation with One-Prompt Meta-Learning (NeurIPS 2024 · similarity 0.89)
- Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.89)
- Adapting Visual-Language Models for Generalizable Anomaly Detection in Medical Images (CVPR 2024 · similarity 0.88)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.88)
- VETime: Vision Enhanced Zero-Shot Time Series Anomaly Detection (NeurIPS 2026 · similarity 0.87)
- Align3D-AD: Cross-Modal Feature Alignment and Dual-Prompt Learning for Zero-shot 3D Anomaly Detection (NeurIPS 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.
