Mapbox is taking its long-standing relationship with OpenStreetMap (OSM) to the next level by transforming traditional 3D map representations from basic, featureless cubes into rich, detailed
procedural buildings. We've developed a programmatic solution that addresses the common issue of missing or inconsistent data in OSM for detailed features like roof types and facades.

In this session, we will reveal how a geospatial computer vision tool combines existing OSM building footprints with high-resolution aerial imagery to automatically classify roof shapes. We will also discuss the logic we use to generate facades based on a building's orientation to adjacent roads.

Attendees will see how these dynamically generated 3D models capture the character of urban architecture , providing subtle cues for navigation and enhancing the map experience without sacrificing performance. We will also touch on the rendering of these models and how they are strategically placed along 3D roads and near 3D landmarks to avoid making the map feel too dense. This approach allows us to create the industry's most true-to-life symbolic representation of the world.