Reproduce “MobileGeo: Exploring Hierarchical Knowledge Distillation for Resource-Efficient Cross-view Drone Geo-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
Jian Sun, Kangdao Liu, Chi Zhang, Chuangquan Chen, Junge Shen, C. L. Philip Chen, Chi-Man Vong
Researchers developed MobileGeo, a framework for cross-view drone geo-localization that achieves high accuracy with low inference costs, enabling deployment on resource-constrained edge devices. The model delivers a Recall@1 of 97.15% on the University-1652 dataset, outperforming prior state-of-the-art methods while using 5.8x fewer FLOPs and achieving 3.3x higher Frames Per Second.
Key insights
- Knowledge distillation can effectively transfer semantic and spatial robustness from complex teacher models to a lightweight student network, making it suitable for edge device deployment without significant performance loss.
- Dynamically adjusting cross-domain feature alignment based on predictive uncertainty helps to mitigate the adverse effects of data imbalance and asymmetric convergence between drone and satellite views.
- Information-theoretic view importance assessment combined with submodular optimization allows for efficient selection and aggregation of a diverse, informative subset of drone views, significantly enhancing multi-view localization accuracy during inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.