Conceptual structure
RoofVIP associates a stable sample identifier with the modalities available in a particular release.
For the multimodal fixed-scene release v1.0.3.0, the conceptual relationship is:
Scene ID
├── RGB
├── DSM
├── DTM
├── nDSM
├── LiDAR point cloud
├── vector roof geometry
└── split / metadata
All products belonging to a scene describe the same spatial extent. Version 1.0.3.0 uses 256 × 256 pixel scenes on a 0.2 m scene grid, corresponding to approximately 51.2 × 51.2 m on the ground.
The original v1.0.0.0 release has a different, primarily object-centred organization. The exact physical folder structure, serialized fields, and modality availability should therefore always be read from the corresponding release page rather than inferred from another version.
Why keep this separate from release pages?
The documentation defines the meaning of common concepts. A release page defines the actual contents of one published dataset version. Keeping those roles separate lets RoofVIP add new representations without rewriting the historical contract of older releases.