Reproduce “ConGeo: Robust Cross-view Geo-localization across Ground View Variations” 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
Li Mi, Chang Xu, Javiera Castillo-Navarro, Syrielle Montariol, Wen Yang, Antoine Bosselut, Devis Tuia
A single model called ConGeo robustly performs cross-view geo-localization for ground images with unknown orientations and limited fields of view, significantly outperforming prior methods on datasets like CVUSA and CVACT while maintaining high accuracy in traditional North-aligned settings.
Key insights
- Explicit contrastive alignment objectives are more effective than simple data augmentation for learning robust feature representations against ground view variations.
- Enforcing proximity between ground view variants and their original or corresponding aerial images, via single-view ground and cross-view transformed contrastive losses, is crucial for improving performance under FoV variations.
- The model learns to focus on semantically consistent objects (e.g., buildings, trees) rather than fragile spatial correspondences, leading to enhanced orientation invariance and generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.