Research map / Detection and segmentation / Open-set and few-shot detection
Omni-AD: A Large-scale and Versatile Benchmark for Industrial Anomaly Detection
Dahu Shi, Chengshen He, Shaochen Zhang, Bo Qian, Xiaochen Quan, Wencong Zhang, Xing Wei
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 |
|---|---|---|---|
| move from MVTec AD to Omni-AD for evaluating Dinomaly | Dinomaly · unsupervised IAD, I-AUROC | I-AUROC: 99.6 -> 84.5 | Hurt |
| move from MVTec AD to Omni-AD for evaluating Dinomaly | Dinomaly · unsupervised IAD, P-AUPRO | P-AUPRO: 94.8 -> 71.4 | Hurt |
| use Omni-AD instead of MVTec AD to separate methods | unsupervised IAD methods · unsupervised IAD, I-AUROC range | I-AUROC range: 94.9~99.9 -> 82.1~87.8 | Improved |
| add one normal reference image (1-shot) for MLLM discrimination (ablation) | Qwen3-VL-30B-A3B-Thinking · Omni-AD defect discrimination (image-level) | accuracy: 62.85 -> 69.38 | Improved |
| add one normal reference image (1-shot) for MLLM classification (ablation) | Qwen3-VL-30B-A3B-Instruct · Omni-AD defect classification (image-level) | accuracy: 62.22 -> 65.94 | Improved |
| add one normal reference image (1-shot) for MLLM localization (ablation) | Qwen3-VL-30B-A3B-Thinking · Omni-AD defect localization | F1: 26.20 -> 25.64 | No clear effect |
| add one normal reference image (1-shot) for LLaVA-NeXT classification (ablation) | LLaVA-NeXT-34B · Omni-AD defect classification (image-level) | accuracy: 38.84 -> 34.80 | Hurt |
| enable thinking mode in Qwen3-VL-30B-A3B | Qwen3-VL-30B-A3B · Omni-AD defect localization (0-shot) | F1: 19.12 -> 26.20 | Improved |
| enable thinking mode in Qwen3-VL-8B | Qwen3-VL-8B · Omni-AD defect discrimination (image-level, 0-shot) | accuracy: 61.81 -> 57.86 | Hurt |
| highlight a specific box region instead of whole image for MLLM questions | MLLMs · Omni-AD discrimination and classification | accuracy: Img-Acc 61.81 -> Box-Acc 68.66 (Qwen3-VL-8B-Instruct, discrimination 0-shot) | Improved |
Closest papers by meaning
- MMAD: A Comprehensive Benchmark for Multimodal Large Language Models in Industrial Anomaly Detection (ICLR 2025 · similarity 0.92)
- AnomalyGPT: Detecting Industrial Anomalies Using Large Vision-Language Models (AAAI 2024 · similarity 0.90)
- Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection (CVPR 2024 · similarity 0.89)
- InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection (NeurIPS 2026 · similarity 0.88)
- Supervised Anomaly Detection for Complex Industrial Images (CVPR 2024 · similarity 0.88)
- Towards Training-free Anomaly Detection with Vision and Language Foundation Models (CVPR 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.
