SYSU-HiRoads: A Large-Scale Multi-Level Road Dataset for Remote Sensing

SYSU-HiRoads is a large-scale, multi-level road dataset designed to support research on road surface segmentation, road network reconstruction, and fine-grained road hierarchy classification from high-resolution remote sensing imagery. The dataset is derived from GF-2 satellite imagery covering Henan Province, China, with a spatial resolution of 0.8 meters. It spans approximately 3,631 km² and consists of 1,079 image tiles, each with a size of 1024 × 1024 pixels. SYSU-HiRoads provides aligned multi-source annotations, including: (1) Pixel-level road masks for dense road surface segmentation; (2) Vector-level road centerlines for topology-preserving network reconstruction; and (3) Three-level hierarchical road labels for fine-grained classification. The dataset bridges the gap between segmentation-oriented and network-oriented road benchmarks by jointly providing raster-level and vector-level representations with hierarchical attributes. SYSU-HiRoads is publicly released to facilitate research in remote sensing, geospatial artificial intelligence, urban studies, and intelligent transportation systems.