RAD
收藏资源简介:
RAD数据集是由华中科技大学创建,专注于评估图像异常检测方法的鲁棒性。该数据集包含自由视角、不均匀光照和模糊采集等条件下的图像,旨在模拟实际工业检测环境。数据集通过使用工业相机采集工作平台上是否存在异物的图像来构建,异物包括螺栓、电缆扎带、海绵和胶带等。创建过程中,考虑了多种图像噪声,如视角变化、光照不均和图像模糊,以确保数据集能有效测试和提升异常检测方法的鲁棒性。RAD数据集主要应用于工业制造中的异常检测,特别是识别工作平台上的异物,以提高生产安全和效率。
The RAD dataset was created by Huazhong University of Science and Technology, focusing on evaluating the robustness of image anomaly detection methods. This dataset includes images captured under conditions such as free viewpoints, non-uniform illumination, and blurry acquisition, aiming to simulate real industrial inspection environments. It is constructed by using industrial cameras to collect images of foreign objects on work platforms, where the foreign objects include bolts, cable ties, sponges, adhesive tapes, and other similar items. During the dataset creation process, multiple types of image noise are considered, including viewpoint changes, uneven illumination, and image blurring, to ensure that the dataset can effectively test and improve the robustness of anomaly detection methods. The RAD dataset is mainly applied to anomaly detection in industrial manufacturing, particularly for identifying foreign objects on work platforms to enhance production safety and efficiency.

- 1RAD: A Comprehensive Dataset for Benchmarking the Robustness of Image Anomaly Detection华中科技大学 · 2024年



