One Query, Many Scales: Sparse Mixture-of-Experts for Efficient Hierarchical Cross-View Geo-Localization

2026-08-02

AI Summary

This work addresses the inefficiency of existing cross-view geolocalization methods in large-scale retrieval, their sensitivity to variations in satellite image resolution, and the error propagation inherent in hierarchical search strategies. To overcome these limitations, we propose GeoMoE, the first approach to introduce sparse Mixture-of-Experts (MoE) into this task. GeoMoE employs a dual-encoder architecture that decouples global multi-scale representation learning from local hierarchical search. By integrating content-adaptive routing and multi-scale supervision, it constructs a resolution-invariant embedding space, while probabilistic beam search substantially reduces computational cost and error accumulation. On the Just Zoom In benchmark, GeoMoE achieves 95.78% R@40m (+2.77%), and on the new VIGOR-M benchmark, it attains 62.39% R@1 with only 0.885 MMAC per query—just 5.27% of the cost of exhaustive L3 scanning—significantly outperforming the strongest baselines.

Abstract

Cross-view geo-localization (CVGL) retrieves geo-tagged satellite imagery for a ground-view query. Most systems exhaustively search a flat, fixed-resolution gallery, incurring high cost over large areas and adapting poorly to satellite resolution changes. Autoregressive coarse-to-fine alternatives reduce comparisons but bind later predictions to earlier decisions and a predefined hierarchy. We introduce GeoMoE, a sparse mixture-of-experts dual encoder that decouples global multi-scale representation learning from local hierarchical search. Global multi-scale supervision and content-adaptive routing map ground and satellite images across resolutions into a globally comparable embedding space. At inference, each image is encoded once, and probabilistic beam search follows parent--child links to score a small candidate subset. Later levels reuse these descriptors rather than features generated by preceding levels, limiting feature-level error propagation and hierarchy coupling. We further introduce VIGOR-M, a four-city benchmark with an explicit parent--child satellite hierarchy and held-out half-step galleries for single-resolution, cross-resolution, and hierarchical evaluation. GeoMoE achieves 95.78% R@40m on Just Zoom In, 2.77 percentage points above the previous best, and 62.39% R@1 on VIGOR-M. The latter requires 0.885 MMAC/query for descriptor matching, 5.27% of an exhaustive L3 scan, while exceeding the strongest exhaustive baseline by 3.12 percentage points in R@1. One model trained on L1, L2, and L3 also outperforms a matched dense control across all six galleries and transfers to three withheld resolutions. By decoupling globally trained embeddings from local hierarchical search, GeoMoE jointly improves localization accuracy, search efficiency, and cross-resolution transfer.

Problem

• cross-view geo-localization
• hierarchical search
• multi-scale representation
• resolution adaptation
• efficient retrieval

Innovation

• Sparse Mixture-of-Experts
• Cross-view Geo-localization
• Hierarchical Search
• Multi-scale Representation
• Content-adaptive Routing