高分辨率带钢表面缺陷数据集CSU_Steel

CSU_STEEL数据集是一个宽幅面、高分辨率板带钢表面缺陷数据集，包含968张大小为1024×4096像素的原始图像数据，这些数据是在复杂的板带钢生产线上采集的，如图1所示为CSU_STEEL数据集的样本示例。每张样本图中包含一种或多种类型的缺陷，如图2所示为缺陷样本局部图。

CSU_STEEL dataset is a wide format and high- resolution dataset of surface defects of plate and strip steel, including 968 pieces of original image with 1024 × 4096 pixels, which are collected on the complex plate and strip production line. Some samples of the STEEL dataset are shown in Fig 1. Each sample image contains one or more types of defects. Partial images of the defect sample are shown in Fig 2.

CSU_STEEL数据集中包含近6500个对象实例，共六类标签，具体为辊印、弹性形变、浪形、夹杂、氧化皮和划痕 。数据集的格式为PASCAL VOC。数据集中实例数量、大小及位置分布情况如图3所示。

CSU_ STEEL dataset contains nearly 6500 object instances, including six types of labels–roll marks, elastic deformation, wave, inclusions, oxide scale and scratches. The format of the dataset is PASCAL VOC. The number, size and location distribution of instances in the dataset are shown in Fig 3.