Reproduce “Geo-Localization via Ground-to-Satellite Cross-View Image Retrieval” 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
Zelong Zeng, Zheng Wang, Fan Yang, Shin'ichi Satoh
Researchers introduced the Peer Learning and Cross Diffusion (PLCD) framework, which leverages drone-view images as an intermediate domain to bridge the gap in ground-to-satellite geo-localization. This framework achieved 40.87% CMC@1% on University-Earth and 21.77% on University-Google datasets, surpassing existing methods by 30.32% and 14.64% respectively in CMC@1%.
Key insights
- Drone-view images effectively act as an intermediate bridge, making ground-to-drone and drone-to-satellite transformations easier to learn than direct ground-to-satellite mapping.
- The peer learning strategy, with its iterative refinement and focus on local features, significantly enhances cross-view matching accuracy between ground and drone images.
- Cross-diffusion is a highly effective method to connect distinct cross-view feature spaces without forcing them into a single, potentially distorting, common space, preserving manifold structures for better ranking.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.