<b>SAFE-Net-Dataset</b>

<b>Dataset Description for SAFE-Net-Dataset</b><b>1. Introduction</b>This dataset was created as part of the research titled <i>"</i> <i>SAFE-Net: Multi-Head Attention Enhanced Framework for Defect Detection in Anti-Corrosion Coatings on Steel Structures"</i>. It focuses on detecting defects commonly found in steel structure coatings, including <b>corrosion, cracks, flaking, and blistering</b>. The goal of this dataset is to facilitate research in computer vision for defect detection, specifically for high-resolution image segmentation and object detection tasks.<b>2. Dataset Overview</b><b>Name</b>: <b>SAFE-Net-Dataset</b><b>Version</b>: 1.0<b>Number of Images</b>: 8,193 images.<b>Defects Categories</b>:Class 0: CorrosionClass 1: CrackClass 2: FlakingClass 3: Blistering<b>Data Formats</b>:<b>Images</b>: RGB images in .jpg format.<b>Annotations</b>: Provided in .txt format, where each .txt file corresponds to an image and contains information about the bounding boxes for all objects in that image.<b>3. Methodology</b><b>3.1. Data Collection</b>The images were collected from industrial steel structures under real-world conditions. Images were captured using high-resolution cameras to ensure fine detail resolution of small defects like microcracks.<b>3.2. Data Annotation</b>Annotation tools used include <b>Labelimg</b> for bounding boxes. Then, the XML file that has been obtained is to be converted into TXT format.<b>4. Citation</b>If you use this dataset, please cite the following paper:<i>[Paper Title:</i> <i>SAFE-Net: Multi-Head Attention Enhanced Framework for Defect Detection in Anti-Corrosion Coatings on Steel Structures]</i><br>Authors: [Yue Yu, Shouchao Jiang, Yijun Wang, Shaojun Zhu*]<b>5. License</b>This dataset is released under the <b>GPL</b><b> </b>license. Users are free to use, modify, and distribute the dataset, provided that appropriate credit is given to the original authors.For questions or issues, please contact [e-mail].<br><b>SAFE-Net-Dataset 数据集说明</b><b>1. 研究背景</b>本数据集依托题为<i>《SAFE-Net：面向钢结构防腐涂层缺陷检测的多头注意力（Multi-Head Attention）增强框架》</i>的研究项目构建而成，聚焦于钢结构防腐涂层的常见缺陷检测任务，涵盖<b>腐蚀、裂纹、起皮与鼓泡</b>四类缺陷。本数据集旨在推动计算机视觉领域的缺陷检测研究，尤其面向高分辨率图像分割与目标检测任务。<b>2. 数据集概览</b><b>数据集名称</b>: <b>SAFE-Net-Dataset</b><b>版本</b>: 1.0<b>图像数量</b>: 共8193张图像。<b>缺陷类别</b>:<br>类别0：腐蚀<br>类别1：裂纹<br>类别2：起皮<br>类别3：鼓泡<b>数据格式</b>:<b>图像文件</b>: 采用RGB色彩空间的.jpg格式图像。<b>标注文件</b>: 以.txt格式存储，每张图像对应一个同名标注文件，包含该图像内所有目标的边界框信息。<b>3. 数据集构建流程</b><b>3.1 数据采集</b>图像采集自真实工业工况下的钢结构场景，使用高分辨率相机拍摄，以确保微裂纹等小型缺陷的细节分辨率。<b>3.2 数据标注</b>标注采用<b>Labelimg</b>工具生成边界框，随后将生成的XML格式标注文件转换为TXT格式。<b>4. 引用声明</b>若您使用本数据集，请引用如下论文：<i>【论文标题：SAFE-Net：面向钢结构防腐涂层缺陷检测的多头注意力（Multi-Head Attention）增强框架】</i><br>作者：[Yue Yu, Shouchao Jiang, Yijun Wang, Shaojun Zhu*]<b>5. 授权协议</b>本数据集采用<b>GPL（GNU通用公共许可证）</b>开源协议发布，使用者可自由使用、修改及分发本数据集，但需注明原作者的原创贡献。如有任何疑问或问题，请联系邮箱：[e-mail].<br>