CPLID Dataset

This dataset is divided into two part:<code>Normal_Insulators</code> contains the normal insulators capture by UAVs. The number of the normal insulator images is <b>600</b>.<code>Defective_Insulators</code> contains the insulators with defect. The number of the defective insulator images is <b>248</b>. Since we don't have too much defective insulators, the data augmentation method is applied. These images are synthesized by following process:Use the algorithm in TVSeg to segment the defective insulator from a small part original images, the segment results are the mask images;Use affine transform to augment the original images and their mask, the augmentation results is a lot of original-mask image pairs;Use these image pairs to train the U-Net;Use the trained U-Net to segment the rest part of images;Attach the insulators in different backgrounds.Both these two directories contain two subdirectories, one called <code>images</code> contains the image files, the other called <code>labels</code> contains the VOC2007 format annotations.The <code>labels</code> of <code>Normal_Insulators</code> contains <b>only</b> the annotations of insulators;The <code>labels</code> of <code>Defective_Insulators</code> contains not only the annotations of insulators but also the annotations of defects which on the insulators.The images is provided by the State Grid Corporation of China, and the dataset is made by WANG Zi-Hao. If you have any question about this dataset, feel free to contact [e-mail].

本数据集分为两个部分：<code>Normal_Insulators</code> 存储由无人机（Unmanned Aerial Vehicle, UAV）拍摄的正常绝缘子图像，共计600张；<code>Defective_Insulators</code> 存储存在缺陷的绝缘子图像，共计248张。由于可获取的缺陷绝缘子样本数量有限，我们采用了数据增强方法以扩充样本。上述缺陷绝缘子图像通过以下流程合成：首先利用TVSeg算法从少量原始图像中分割出缺陷绝缘子，所得结果为掩码图像；随后对原始图像及其对应的掩码进行仿射变换以完成数据增强，生成大量原始图像-掩码图像对；利用这些图像对训练U-Net模型；再使用训练完成的U-Net对剩余图像进行分割；最后将分割得到的绝缘子附着至不同背景场景中。上述两个数据集目录均包含两个子目录：名为<code>images</code>的子目录存储图像文件，名为<code>labels</code>的子目录存储VOC2007格式的标注文件。其中，<code>Normal_Insulators</code>的<code>labels</code>仅包含绝缘子的标注信息；而<code>Defective_Insulators</code>的<code>labels</code>不仅包含绝缘子的标注信息，还包含绝缘子表面缺陷的标注信息。本数据集的图像由国家电网有限公司提供，由王梓浩（WANG Zi-Hao）整理制作。若您对本数据集有任何疑问，可联系邮箱[e-mail]。