Region Context from Unifying Points, Lines, and Polygons

We envision a new generation of urban foundation models that unify semantic, spatial, and topological relationships within and across point, line, and polygon features. Existing approaches typically encode different feature types separately and fuse them only at later stages, missing opportunities for truly integrated contextualization. To illustrate our vision, we present RegionContext, a framework that generates contextual region embeddings capturing both fine-grained spatial structures and higher-order semantics. Finally, we highlight key research directions for region contextualization in urban foundation models.