A Unified Hierarchical Framework for Fine-grained Cross-view Geo-localization over Large-scale Scenarios

TL;DR

UnifyGeo is a unified hierarchical framework that integrates retrieval and metric localization for large-scale, fine-grained geo-localization, significantly improving accuracy and efficiency over previous methods.

Contribution

The paper introduces UnifyGeo, a novel framework that jointly learns multi-granularity representations and employs a re-ranking mechanism, unifying two key localization tasks into one network.

Findings

Substantially outperforms state-of-the-art methods on VIGOR benchmark.

Improves 1-meter localization recall from 1.53% to 39.64%.

Enhances cross-area localization recall from 0.43% to 25.58%.

Abstract

Cross-view geo-localization is a promising solution for large-scale localization problems, requiring the sequential execution of retrieval and metric localization tasks to achieve fine-grained predictions. However, existing methods typically focus on designing standalone models for these two tasks, resulting in inefficient collaboration and increased training overhead. In this paper, we propose UnifyGeo, a novel unified hierarchical geo-localization framework that integrates retrieval and metric localization tasks into a single network. Specifically, we first employ a unified learning strategy with shared parameters to jointly learn multi-granularity representation, facilitating mutual reinforcement between these two tasks. Subsequently, we design a re-ranking mechanism guided by a dedicated loss function, which enhances geo-localization performance by improving both retrieval accuracy and metric localization references. Extensive experiments demonstrate that UnifyGeo significantly outperforms the state-of-the-arts in both task-isolated and task-associated settings. Remarkably, on the challenging VIGOR benchmark, which supports fine-grained localization evaluation, the 1-meter-level localization recall rate improves from 1.53\% to 39.64\% and from 0.43\% to 25.58\% under same-area and cross-area evaluations, respectively. Code will be made publicly available.