6,300 high-resolution panoramic images aligned with a 3D CAD model of London, covering 20 square kilometers of urban environment. The dataset provides dense annotations for surface normals, depth, and semantic segmentation across 12 distinct urban categories.
Use Cases
- Train monocular depth estimation models using the provided depth maps and panoramic images
- Develop semantic segmentation algorithms for urban scenes using the 12 category labels
- Evaluate 3D reconstruction techniques by comparing predicted geometry against the aligned CAD model
Strengths
- 6,300 high-resolution street-view panoramas aligned with 3D geometry
- Covers a 20 km² area of London with structural annotations
- Includes ground truth for surface normals, depth maps, and semantic segmentation
- Features 12 semantic categories for urban infrastructure and building components