Reproduce “Fusing Satellite Imagery and Planimetric Maps for Cross-View Localization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Quang Long Ho Ngo, Zimin Xia, Alexandre Alahi
EPFL researchers introduce a plug-in fusion module that combines satellite imagery with OSM-derived planimetric maps for ground-to-bird’s-eye-view localization. Cross-modal deformable attention and patch-wise adaptive weighting let the system favor visual or geometric evidence by region and scale, reducing CCVPE’s KITTI cross-area mean position error from 9.16 m to 3.85 m and improving results across multiple localization architectures.
Key insights
- The most useful modality varies by scene: satellite imagery can distinguish visual structures, while OSM geometry is more reliable where vegetation or overhead occlusion obscures the ground layout.
- Context-aware processing and adaptive patch fusion are complementary; on VIGOR, they reduce CCVPE’s mean error from 3.60 m to 3.10 m and 3.23 m individually, and to 2.87 m together.
- The learned fusion generally combines both modalities instead of selecting one globally, with weights changing across regions and scales; this supports better transfer across geographic areas.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.