Texture2LoD3: Enabling LoD3 Building Reconstruction With Panoramic Images

This dataset is designed to support high-fidelity 3D reconstruction of urban facades (LoD3) by integrating geometric, semantic, and texture information. Its structure is as follows:

• citygml & citygml_buildings.geojson: Provide spatial information, including city-wide building footprints and associated metadata.
• obj: Contains semantically segmented WallSurfaces, which serve as the basis for further texturing tasks.
• panoramas: Comprises panorama images sourced from Google Street View, offering rich visual context for texture rectification and mapping.
• gt_masks: Offers ground truth masks of LoD3 facades, which are used for training and evaluating segmentation approaches. They can also be obtained by our open-source tool: LoD3Masks
• textures: Consists of facade textures extracted using our Texture2LoD3 method from panoramic images, ensuring high-quality surface visual fidelity for 3D models.

By combining multiple modalities, this dataset provides a comprehensive foundation for urban facade modeling, texture mapping, and related research. This dataset also contributes to the TUM2TWIN project.