RMDCrack：A Road Multi-Dimensional Crack Dataset

RMDCrack: A Road Multi-Dimensional Crack Dataset is a comprehensive dataset designed for intelligent pavement crack detection and multi-modal surface analysis. It consists of 3,200 samples across 2D, 2.5D, and 3D modalities, providing rich information for both texture-based and geometry-aware crack understanding. The 2D subset includes 1,500 highway images captured by a multifunctional road inspection vehicle—comprising 922 crack images and 578 easily misclassified images—as well as 900 additional crack images collected under challenging conditions such as shadows, stains, watermarks, and rutting. The 2.5D subset contains 407 sets of aligned RGB images, depth maps, and surface normal maps, offering enhanced structural cues for robust crack perception. The 3D subset provides 407 road-surface point clouds, enabling the reconstruction and analysis of cracks in true spatial geometry. Together, RMDCRack offers a multi-source, multi-modal, and multi-scale benchmark that supports a wide range of tasks, including crack detection, semantic segmentation, 3D reconstruction, and multi-modal fusion.