GroundControl: Satellite-Conditioned Multi-View Diffusion for Reconstruction-Free Novel View Synthesis
Satellite imagery helps keep generated street-level views aligned with the scene’s actual layout.
GroundControl combines satellite imagery with a handful of posed street-level images to generate new outdoor views without reconstructing a 3D scene. Its cross-attention conditioning preserves pretrained diffusion capabilities while using the satellite view to constrain spatial layout. The authors report better results than satellite-only and street-view-only baselines, especially when target views barely overlap the source images. The paper is accepted to WACV 2027.
ArXiv · AI/CL/LG's note
GroundControl combines satellite imagery with a handful of posed street-level images to generate new outdoor views without reconstructing a 3D scene. Its cross-attention conditioning preserves pretrained diffusion capabilities while using the satellite view to constrain spatial layout. The authors report better results than satellite-only and street-view-only baselines, especially when target views barely overlap the source images. The paper is accepted to WACV 2027.
ArXiv · AI/CL/LG's note
score 5