Texture-AD
收藏资源简介:
Texture-AD数据集由南京大学苏州集成电路学院和未来智能芯片交叉研究中心创建,专注于工业质量检测中的无监督异常检测。该数据集包含43120张高分辨率图像,涵盖15种不同布料、14种半导体晶圆和10种金属板,模拟真实工业检测场景。数据集包括多种类型的缺陷,如划痕、皱纹、颜色变化和点缺陷,并提供像素级标注。创建过程中,图像通过高分辨率工业相机采集,并使用Labelme工具进行人工标注。Texture-AD旨在评估和提升工业缺陷检测算法的鲁棒性和泛化能力,解决现有数据集与实际生产环境数据差异大的问题。
The Texture-AD dataset was developed by the School of Integrated Circuits (Suzhou Campus) of Nanjing University and the Cross-disciplinary Research Center for Future Intelligent Chips, focusing on unsupervised anomaly detection in industrial quality inspection. It contains 43,120 high-resolution images spanning 15 fabric categories, 14 semiconductor wafer categories and 10 metal sheet categories, simulating realistic industrial inspection scenarios. The dataset includes diverse defect types such as scratches, wrinkles, color variations and point defects, with pixel-level annotations provided. During the dataset construction, images were collected via high-resolution industrial cameras and manually annotated using the Labelme tool. Texture-AD aims to evaluate and improve the robustness and generalization capability of industrial defect detection algorithms, solving the problem of large discrepancies between existing datasets and real production environment data.



