Dual datasets for anomaly detection on metal surface under the influence of impurities
收藏资源简介:
该数据集由弗劳恩霍夫工业数学研究所(ITWM)创建,旨在解决金属表面在杂质(如水渍、指纹、贴纸等)影响下的异常检测问题。数据集包含高分辨率图像,模拟了真实工业环境中的表面杂质,并生成了对应的合成数据。数据集通过程序化方法生成,使用Perlin噪声模拟水渍的自然形态,并通过抖动采样方法控制杂质的分布。该数据集主要用于训练和评估异常检测模型,特别是在高分辨率图像处理中的内存瓶颈问题。通过该数据集,研究人员可以更好地理解杂质对模型性能的影响,并优化模型在实际工业场景中的应用。
This dataset was created by the Fraunhofer Institute for Industrial Mathematics (ITWM) to address the anomaly detection problem of metal surfaces affected by impurities such as water stains, fingerprints, stickers, and others. The dataset includes high-resolution images that simulate surface impurities in real industrial environments, alongside corresponding synthetic data. It is generated via a programmatic approach: Perlin noise is used to simulate the natural morphology of water stains, while jittered sampling is employed to control the distribution of impurities. This dataset is primarily used for training and evaluating anomaly detection models, especially to tackle memory bottleneck issues encountered in high-resolution image processing. With this dataset, researchers can better understand the impact of impurities on model performance and optimize the practical industrial application of these models.

- 1Sequential PatchCore: Anomaly Detection for Surface Inspection using Synthetic Impurities弗劳恩霍夫工业数学研究所(ITWM) · 2025年



