Reproduce “CLNet: Cross-View Correspondence Makes a Stronger Geo-Localizationer” 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
Xianwei Cao, Dou Quan, Shuang Wang, Ning Huyan, Wei Wang, Yunan Li, Licheng Jiao
CLNet, a framework for image retrieval-based cross-view geo-localization, explicitly models cross-view correspondences to bridge the semantic and geometric gaps between ground and satellite images. It achieves Recall@1 scores of 98.77% on CVUSA and 71.89% on the CVACT test set, outperforming prior methods while reducing computational complexity by 68.2% in FLOPs.
Key insights
- Explicitly modeling cross-view correspondences through learnable interaction mechanisms can enhance geo-localization accuracy and robustness more effectively than independent feature learning.
- A lightweight, data-driven view conversion module (NEC) can implicitly bridge geometric gaps between vastly different perspectives without requiring precise camera pose information.
- Recalibrating features using correspondence-aware neural maps (GFR) improves view consistency and generalizability while maintaining computational efficiency, ensuring practical applicability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.