HighRPD

In order to meet the data needs for road pavement distress detection, we have created a standardized dataset road pavement distress named HighRPD, which consists of road pavement distress images captured from a drone perspective. This dataset maintains a uniform image resolution of 640x640 pixels, in alignment with the test dataset specifications for the YOLO v8 model. Meanwhile, informed by the prevalence of road pavement distress, the dataset HighRPD specifically targets road pavement distress classified into three fundamental categories: line, block, and pit. We utilized the Labelbox platform in combination with DarkLabel for constructing our dataset. In summary, a total of 11,696 road pavement images were successfully labeled, including 12,365 line annotations, 8,239 block annotations, and 1,412 pit annotations.
The HighRPD dataset comprises two subfolders: one named 'images' and the other 'labels'. The 'images' folder contains pictures sized 640x640 pixels in JPG format, while the 'labels' folder contains txt files with labels formatted in the YOLO style. Each object is represented by a single line, formatted as 'class center_x center_y width height'. There are three classes: class 0 for lines, class 1 for blocks, and class 2 for pits. The coordinates (x_center, y_center, width, height) are normalized by dividing x_center and width by the image width, and y_center and height by the image height.

为满足道路路面病害检测的数据需求，我们构建了一个名为HighRPD的标准数据集，该数据集由无人机视角捕获的道路路面病害图像组成。该数据集保持了与YOLO v8模型测试数据集规范一致的640x640像素的统一图像分辨率。鉴于道路路面病害的普遍性，HighRPD数据集特别针对道路路面病害的三种基本类别进行分类：线状、块状和坑洼。我们结合Labelbox平台与DarkLabel工具构建了我们的数据集。简而言之，共成功标注了11,696幅道路路面图像，包括12,365条线状标注、8,239块状标注和1,412坑洼标注。HighRPD数据集包含两个子目录：一个名为'images'，另一个名为'labels'。'images'目录包含640x640像素大小的JPG格式图片，而'labels'目录包含以YOLO风格格式化的txt标签文件。每个对象由一行表示，格式为'class center_x center_y width height'。共有三个类别：类别0代表线条，类别1代表块状，类别2代表坑洼。坐标（x_center, y_center, width, height）通过将x_center和width除以图像宽度，将y_center和height除以图像高度进行归一化处理。

搜集汇总

背景与挑战

背景概述

HighRPD是一个用于道路路面病害检测的标准化数据集，包含从无人机视角采集的11,696张640x640像素图像，标注了线状、块状和坑洼三类病害，共22,016个标注实例。数据集采用YOLO格式标注，专为YOLO v8模型优化，适用于目标检测任务，由西南交通大学创建并共享。