Research map / 3D vision / Depth and stereo
Georeferenced Cross-View 3D Reconstruction from Ground and Satellite Images
David G Shatwell, Mubarak Shah
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a method for georeferenced 3D reconstruction that jointly uses ground images and satellite imagery to place camera centers and 3D points in a geographic frame (abstract only; details unavailable).
Problem
Georeferenced 3D reconstruction from ground images requires recovering local scene geometry and localizing every camera and 3D point in geographic coordinates; existing cross-view geo-localization estimates only single camera GPS without dense geometry, and feed-forward multi-view reconstruction mod
Approach
Combines feed-forward multi-view 3D reconstruction of ground images with cross-view satellite imagery so that the reconstruction is georeferenced in a geographic coordinate frame (exact mechanism not stated in the abstract).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| joint ground-satellite tokens with satellite alignment head instead of VGGT plus Loc2 alignment | X-VGGT · KITTI pixel-wise geo-registration | geo-registration error (m): 178.45 -> 25.52 | Improved |
| fit reconstructed camera centers to Loc2 predictions (geo-localization-aligned reconstruction) | VGGT · KITTI pixel-wise geo-registration | geo-registration error (m): 146.35 -> 178.45 (VGGT vs VGGT+Loc2) | Hurt |
| joint encoding of ground and satellite instead of late fusion (ablation) | X-VGGT · KITTI camera geo-localization | orientation error (deg): 95.07 -> 22.50 | Improved |
| add georeferencing supervision on top of VGGT | VGGT · Metropolis local 3D reconstruction | AUC: 55.31 -> 51.34 | Hurt |
| add georeferencing supervision on top of VGGT | VGGT · WRIVA local 3D reconstruction | AUC: 16.90 -> 21.17 | Improved |
| add dense projection and camera-center losses to parameter loss (ablation) | X-VGGT · KITTI | geo-registration and localization error | Improved |
| increase number of ground images (ablation) | X-VGGT · Metropolis | localization and orientation error | Improved |
| downsample satellite image resolution to 130x130 (ablation) | X-VGGT · KITTI (512 scenes) | localization error (m): 12.88 -> 25.28 | Hurt |
| shift satellite image by 10 m (misalignment) (ablation) | X-VGGT · KITTI (512 scenes) | localization error (m): 12.88 -> 16.74 | Hurt |
| larger satellite crop (187.5 m vs 150 m) (ablation) | X-VGGT · KITTI (512 scenes) | localization error (m): 12.88 -> 17.06 | Hurt |
| replace predicted alignment with oracle alignment from GT camera locations (ablation) | X-VGGT · Metropolis pixel-wise geo-registration | geo-registration error (m): 17.84 -> 16.84 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 3 of 4 shared results
- Loc2 — Loc$^{2}$: Interpretable Cross-View Localization via Depth-Lifted Local Feature Matching (ICLR 2026) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Geo2: Geometry-Guided Cross-view Geo-Localization and Image Synthesis (CVPR 2026 · similarity 0.90)
- Boosting 3-DoF Ground-to-Satellite Camera Localization Accuracy via Geometry-Guided Cross-View Transformer (ICCV 2023 · similarity 0.89)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.88)
- Cross-View Splatter: Feed-Forward View Synthesis with Georeferenced Images (CVPR 2026 · similarity 0.88)
- Aligning Geometric Spatial Layout in Cross-View Geo-Localization via Feature Recombination (AAAI 2024 · similarity 0.87)
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
