21 high-resolution satellite images covering urban areas with pixel-level annotations for three distinct road-related tasks. The dataset includes ground truth masks for road surfaces, road edges, and road centerlines to facilitate multi-task learning in geoscience applications.
Use Cases
- Train multi-task deep learning models to jointly predict road surfaces and edges
- Extract topological road networks using the centerline and surface ground truth
- Improve boundary accuracy in segmentation tasks by supervising models with the road edge labels
- Benchmark the performance of remote sensing architectures on urban road extraction tasks
Strengths
- Contains pixel-level binary masks for road surface segmentation
- Includes specific ground truth labels for road edge detection
- Provides road centerline annotations to support network topology extraction
- Covers 21 high-resolution images of urban environments