HighRPD
Description
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.
Files
Steps to reproduce
We kept the drone's altitude at a consistent 50 meters above the ground during the data collection process. The road pavement images were captured at a resolution of 8192*5460 pixels. Specifically, we screened a total of 546 original large images taken by UAVs. We uniformly cropped these images into multiple small images of 640*640 pixels, adding black edges to areas where the images were too small to fit the desired size. Through several rounds of data annotation and verification, a total of 11,696 road pavement images were successfully labeled.