New Geo² framework unifies geo-localization and image synthesis by leveraging geometric foundation models to bridge the viewpoint gap between ground and aerial imagery

Geo² is a unified framework that combines Cross-View Geo-Localization (CVGL) and Cross-View Image Synthesis (CVIS) tasks using geometric priors from foundation models like VGGT

The framework introduces GeoMap, which embeds ground and aerial features into a shared 3D-aware latent space to reduce cross-view discrepancies

Leverages recent Geometric Foundation Models (GFMs) that extract generalizable 3D geometric features from images for improved cross-view geo-spatial learning

Addresses the challenge of large viewpoint gaps between ground-level and aerial imagery that previously made direct application of 3D reconstruction models difficult