PaveCrack1300: A UAV-Acquired Pavement Crack Segmentation Dataset

Description

PaveCrack1300 is a pixel-level annotated pavement crack dataset comprising 1,300 image–mask pairs acquired by a DJI Mini 4 Pro UAV over road surfaces at Shenyang Jianzhu University and the surrounding road network, Shenyang, China. Images are 512 × 512 pixels (JPEG, RGB). Binary PNG masks encode crack pixels as 255 and background as 0. Mean crack pixel occupancy is 10.80% (median 8.61%, range 4.03–56.14%). The dataset supports training and benchmarking of segmentation models for automated pavement crack detection.

Steps to reproduce

Raw aerial images were orthorectified to a near-nadir viewpoint and cropped into non-overlapping 512×512 px patches. Patches with visible crack content were retained and manually annotated using LabelMe in polygon mode. Annotations were converted to binary PNG masks via a custom Python script. Quality control removed patches with crack pixel occupancy below 4%.

Institutions

• Shenyang Jianzhu University

  Liaoning, Shenyang

Categories

Funders

• National Key R&D Program of China

  Grant ID: Grant No. 2024YFC38098 and 2024YFC3809803
• Liaoning Xingliao Talents Program for Science and Technology Innovation Team of China

  Grant ID: No. XLYC2404005
• Technology Research and Development Program of Shenyang Science and Technology Bureau

  Grant ID: Grant No. 24-213-3-33

Licence