SUT-Crack: A comprehensive dataset for pavement crack detection across all methods

The SUT-Crack dataset (Sharif University of Technology Crack Dataset) presents a collection of high-quality images depicting asphalt pavement cracks specifically designed to facilitate crack detection using various deep learning methods, including classification, object detection, segmentation, etc. During the dataset creation process, careful consideration was given to encompass all possible crack detection challenges, such as the presence of oil stains and shadows on the pavement surface along with varying lighting conditions. The dataset comprises 130 images designed specifically for segmentation and object detection tasks. Each image is accompanied by precise ground truth annotations. This dataset is well-suited for various crack detection methods, offering accurate annotations that enhance its reliability and usefulness across diverse applications. Moreover, the images were taken from a fixed height of 672 mm above the pavement surface, enabling straightforward calibration to derive real-world crack lengths from pixel measurements. A notable feature of the SUT-Crack dataset is the inclusion of geotags, affixing each image with precise latitude and longitude coordinates. This geotagging capability allows for the visualization of the images on a map and imparting valuable geographical context to the dataset. Additionally, by dividing the original images into 200×200 pixel images, over 25,000 images were produced and then categorized into “with crack” and “without crack” classes which can be used for classification purposes. SUT-Crack is available at https://doi.org/10.17632/gsbmknrhkv.6.