Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

(MGS)$^2$-Net: Unifying Micro-Geometric Scale and Macro-Geometric Structure for Cross-View Geo-Localization

Minglei Li, Mengfan He, Chao Chen, Ziyang Meng

TL;DR

This paper addresses cross-view geo-localization under severe 3D geometric misalignment by introducing a geometry-grounded framework, (MGS)², that couples Macro-Geometric Structure Filtering (MGSF) with Micro-Geometric Scale Adaptation (MGSA) and a Geometric-Appearance Contrastive Distillation (GACD) loss. MGSA uses depth priors to dynamically fuse multi-scale features, while MGSF uses depth-derived macro-gradients and normal clustering to suppress view-dependent vertical facades and emphasize horizontal planes. GACD enforces a geometric-prior-based ranking, rewarding roof-centric activations over facade distractions, and the optimization combines L_triplet with L_GACD. Experiments on University-1652 and SUES-200 show state-of-the-art Recall@1 and strong cross-dataset generalization, with qualitative visualizations confirming effective suppression of vertical artifacts. Overall, the approach demonstrates that explicit 3D structure priors substantially improve cross-view localization in urban environments and offers a path toward robust, geometry-aware CVGL systems in GNSS-denied scenarios.

Abstract

Cross-view geo-localization (CVGL) is pivotal for GNSS-denied UAV navigation but remains brittle under the drastic geometric misalignment between oblique aerial views and orthographic satellite references. Existing methods predominantly operate within a 2D manifold, neglecting the underlying 3D geometry where view-dependent vertical facades (macro-structure) and scale variations (micro-scale) severely corrupt feature alignment. To bridge this gap, we propose (MGS)$^2$, a geometry-grounded framework. The core of our innovation is the Macro-Geometric Structure Filtering (MGSF) module. Unlike pixel-wise matching sensitive to noise, MGSF leverages dilated geometric gradients to physically filter out high-frequency facade artifacts while enhancing the view-invariant horizontal plane, directly addressing the domain shift. To guarantee robust input for this structural filtering, we explicitly incorporate a Micro-Geometric Scale Adaptation (MGSA) module. MGSA utilizes depth priors to dynamically rectify scale discrepancies via multi-branch feature fusion. Furthermore, a Geometric-Appearance Contrastive Distillation (GACD) loss is designed to strictly discriminate against oblique occlusions. Extensive experiments demonstrate that (MGS)$^2$ achieves state-of-the-art performance, recording a Recall@1 of 97.5\% on University-1652 and 97.02\% on SUES-200. Furthermore, the framework exhibits superior cross-dataset generalization against geometric ambiguity. The code is available at: \href{https://github.com/GabrielLi1473/MGS-Net}{https://github.com/GabrielLi1473/MGS-Net}.

(MGS)$^2$-Net: Unifying Micro-Geometric Scale and Macro-Geometric Structure for Cross-View Geo-Localization

TL;DR

This paper addresses cross-view geo-localization under severe 3D geometric misalignment by introducing a geometry-grounded framework, (MGS)², that couples Macro-Geometric Structure Filtering (MGSF) with Micro-Geometric Scale Adaptation (MGSA) and a Geometric-Appearance Contrastive Distillation (GACD) loss. MGSA uses depth priors to dynamically fuse multi-scale features, while MGSF uses depth-derived macro-gradients and normal clustering to suppress view-dependent vertical facades and emphasize horizontal planes. GACD enforces a geometric-prior-based ranking, rewarding roof-centric activations over facade distractions, and the optimization combines L_triplet with L_GACD. Experiments on University-1652 and SUES-200 show state-of-the-art Recall@1 and strong cross-dataset generalization, with qualitative visualizations confirming effective suppression of vertical artifacts. Overall, the approach demonstrates that explicit 3D structure priors substantially improve cross-view localization in urban environments and offers a path toward robust, geometry-aware CVGL systems in GNSS-denied scenarios.

Abstract

Cross-view geo-localization (CVGL) is pivotal for GNSS-denied UAV navigation but remains brittle under the drastic geometric misalignment between oblique aerial views and orthographic satellite references. Existing methods predominantly operate within a 2D manifold, neglecting the underlying 3D geometry where view-dependent vertical facades (macro-structure) and scale variations (micro-scale) severely corrupt feature alignment. To bridge this gap, we propose (MGS), a geometry-grounded framework. The core of our innovation is the Macro-Geometric Structure Filtering (MGSF) module. Unlike pixel-wise matching sensitive to noise, MGSF leverages dilated geometric gradients to physically filter out high-frequency facade artifacts while enhancing the view-invariant horizontal plane, directly addressing the domain shift. To guarantee robust input for this structural filtering, we explicitly incorporate a Micro-Geometric Scale Adaptation (MGSA) module. MGSA utilizes depth priors to dynamically rectify scale discrepancies via multi-branch feature fusion. Furthermore, a Geometric-Appearance Contrastive Distillation (GACD) loss is designed to strictly discriminate against oblique occlusions. Extensive experiments demonstrate that (MGS) achieves state-of-the-art performance, recording a Recall@1 of 97.5\% on University-1652 and 97.02\% on SUES-200. Furthermore, the framework exhibits superior cross-dataset generalization against geometric ambiguity. The code is available at: \href{https://github.com/GabrielLi1473/MGS-Net}{https://github.com/GabrielLi1473/MGS-Net}.
Paper Structure (33 sections, 15 equations, 4 figures, 4 tables)

This paper contains 33 sections, 15 equations, 4 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Work
  3. Foundation Models and Cross-View Geo-Localization
  4. Fine-Grained Alignment and Semantic Fusion
  5. Geometry-Aware Cross-View Geo-Localization
  6. Methodology
  7. Overview
  8. Micro-Geometric Scale Adaptation (MGSA)
  9. Depth-Aware Scale Fusion (DASF)
  10. Macro-Geometric Structure Filtering (MGSF)
  11. Depth-to-Feature Alignment
  12. Macro Gradient via Dilated Convolution
  13. Dominant Plane Perception via Normal Clustering
  14. Adaptive Gating with Neutral Edge Rectification
  15. Residual Geometric Modulation
  16. ...and 18 more sections

Figures (4)

  • Figure 1: From Texture Dependency to Geometric Grounding. (a) Visual Ambiguity: Existing methods often overfit to view-dependent vertical facades (red boxes) that are invisible in the satellite orthophoto, leading to retrieval failure. (b) Our Solution: By explicitly modeling 3D macro-geometric structures, our MGS-Net filters out these vertical artifacts via MGSF and robustly focuses on view-invariant rooftops (green boxes), ensuring consistent cross-view alignment. The image materials are from the University-1652 dataset.
  • Figure 2: The overall framework of the proposed (MGS)². It mainly consists of the Micro-Geometric Scale Adaptation (MGSA) module to handle scale variations and the Macro-Geometric Structure Filtering (MGSF) module to suppress view-dependent artifacts.
  • Figure 3: Visualization of the Macro-Geometric Structure Filtering (MGSF) Mechanism. We visualize the feature response maps before and after applying MGSF. (a) Original Features: The backbone naively activates on high-frequency vertical textures. (b) Enhanced Features: MGSF redistributes attention based on geometric priors. Note that black boxes indicate the effective suppression of view-dependent vertical facades, while white boxes highlight the adaptive enhancement of view-invariant Horizontal planes.
  • Figure 4: Qualitative image retrieval results for the University-1652 dataset. (Top) Top 5 retrieval results for target localization in drone view. (Bottom) Top 5 retrieval results for drone navigation. The first and third rows show retrieval results for (MGS)²; the second and fourth rows show retrieval results for the baseline (DINOv2 + SALAD). Green boxes indicate correctly-matched images; red boxes indicate incorrectly-matched images.