Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection

Real-IAD: A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection

19 Mar 2024 | Chengjie Wang, Wenbing Zhu, Bin-Bin Gao, Zhenye Gan, Jiangning Zhang, Zhihao Gu, Shuguang Qian, Mingang Chen, Lizhuang Ma
Real-IAD is a large-scale, real-world, multi-view industrial anomaly detection dataset containing 150,000 high-resolution images of 30 different objects, significantly larger than existing datasets. It features a larger range of defect areas and ratios, making it more challenging. The dataset was created using a multi-view shooting method and includes sample-level evaluation metrics. It also introduces a new setting for fully unsupervised industrial anomaly detection (FUIAD), based on the observation that industrial production yield rates are typically over 60%, making it more practical. The dataset includes images of various materials and objects from multiple angles to capture subtle defects. It provides high-resolution images and detailed annotations, supporting multi-view analysis. The dataset was evaluated against popular IAD methods, showing that it presents a significant challenge, thus promoting the development of more effective anomaly detection algorithms. Real-IAD aims to bridge the gap between research and real-world applications by providing a comprehensive benchmark for industrial anomaly detection.Real-IAD is a large-scale, real-world, multi-view industrial anomaly detection dataset containing 150,000 high-resolution images of 30 different objects, significantly larger than existing datasets. It features a larger range of defect areas and ratios, making it more challenging. The dataset was created using a multi-view shooting method and includes sample-level evaluation metrics. It also introduces a new setting for fully unsupervised industrial anomaly detection (FUIAD), based on the observation that industrial production yield rates are typically over 60%, making it more practical. The dataset includes images of various materials and objects from multiple angles to capture subtle defects. It provides high-resolution images and detailed annotations, supporting multi-view analysis. The dataset was evaluated against popular IAD methods, showing that it presents a significant challenge, thus promoting the development of more effective anomaly detection algorithms. Real-IAD aims to bridge the gap between research and real-world applications by providing a comprehensive benchmark for industrial anomaly detection.
Reach us at info@study.space
[slides] Real-IAD%3A A Real-World Multi-View Dataset for Benchmarking Versatile Industrial Anomaly Detection | StudySpace