A new interactive 3D diorama of central Warsaw provides a stylized, voxel-based representation of the city center. The project, which covers a 3950 x 3150 m area, renders the city at 1-meter voxel resolution in a style inspired by MagicaVoxel. It includes major landmarks such as the Palace of Culture and Science (PKiN), the Złota 44 tower, and the Vistula boulevards. The map is anchored on the PKiN spire and rotated to align with the Palace axes, ensuring that street grids and building walls appear correctly without 'staircasing' artifacts.
The project's core mechanism involves fusing multiple types of public geographic data. It utilizes LiDAR surface models and terrain models from the Polish Geodetic and Cartographic Office (GUGiK) to establish building heights, roof shapes, and ground relief. This physical structure is then layered with OpenStreetMap (OSM) data, which provides facade colors, materials, road surfaces, and information on over 3,000 trees. To handle discrepancies between these sources—such as when LiDAR under-samples thin spires or fails to see new construction—the system uses a cross-check pipeline. For example, if a building's height in OSM is greater than the LiDAR-captured height, the system assumes the building was constructed after the 2025 flight and prioritizes the OSM data.
To ensure the map remains performant in a web browser, the city is organized into a quadtree of tiles. The viewer initially loads the map at a coarse 8-meter resolution and progressively refines the detail to 1-meter voxels only for the area immediately surrounding the user's position. This streaming approach significantly optimizes resource usage; compared to an all-in-memory model, the quadtree system reduced the JavaScript heap memory from approximately 280 MB to a range of 90–230 MB and improved the initial load time to under 3 seconds. However, the model has specific limitations: windows are procedurally generated rather than mapped individually, and certain structures—such as the Roma and Lilium towers—are excluded because they were not yet built during the 2025 data collection. Additionally, while trams are animated along real track geometry, their specific positions are not synchronized with real-time data.