Sat3DGen: Comprehensive Street-Level 3D Scene Generation from Single Satellite Image

Researchers introduced Sat3DGen, a geometry-first feed-forward framework for generating street-level 3D scenes from a single satellite image.

The method adds new geometric constraints and perspective-view training to improve both reconstruction accuracy and photorealism.

A benchmark built with VIGOR-OOD and high-resolution DSM data shows better RMSE and FID than prior approaches.

The work also demonstrates downstream uses in meshing, video generation, and DSM estimation, highlighting strong satellite-to-street synthesis.