Cross-view geo-localization remains challenging due to drastic appearance variations and geometric distortions caused by viewpoint changes. Existing methods primarily rely on discriminative feature learning or explicit geometric alignment, yet often overlook an important observation: humans tend to match scenes by integrating the spatial context of multiple landmarks rather than relying solely on isolated local visual cues. Inspired by this, we propose a Spatial Context Modeling (SCM) module, which traverses feature maps along four orthogonal directions and models interactions among the resulting directional representations at each spatial position. By aggregating complementary directional observations, SCM constructs richer scene layout representations. Since reliable spatial context depends on robust visual features, we introduce a High-Frequency Information Guidance (HFIG) module that extracts stable structural cues from the frequency domain. By emphasizing edges and contours that are less sensitive to appearance variations, HFIG encourages the model to learn more robust representations. Extensive experiments demonstrate that our method achieves competitive performance on standard cross-view geo-localization benchmarks while exhibiting strong robustness under more challenging scenarios with unknown orientations and limited fields of view (FoVs).
Figures & tables
Figure 1: Illustration of the motivation behind our method. Existing methods mainly rely on discriminative appearance features, which will lead to misalignment due to view shifts. When viewpoints change, the relative spatial arrangement of scene elements provides more stable cues than isolated appearance features. Our method aggregates complementary observations from four scanning directions to build richer scene layout representations for robust cross-view alignment.
Figure 2: Overview of the proposed method. Illustration of the High-frequency Information Guidance (HFIG) and Spatial Context Modeling (SCM). The HFIG module converts the image into the frequency domain using DCT and then extracts fine-grained information of high-frequency. SCM captures the holistic spatial layout via position-wise multi-head attention over four directional scanning paths.
Method
CVUSA
R@1
R@5
R@10
R@1%
SAFA †
89.84
96.93
98.14
99.64
TransGeo
94.08
98.36
99.04
99.77
GeoDTR †
95.43
98.86
99.34
99.86
PaSS-KD †
94.09
98.42
99.13
99.77
Sample4Geo
98.68
99.68
99.78
99.87
Table 1: Performance comparison on CVUSA dataset. Best results are in bold, second-best in underlined. “-” means the score is not provided in the original paper.
Method
CVACT_val
CVACT_test
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
SAFA †
81.03
92.80
94.84
98.17
55.50
79.94
85.08
94.49
TransGeo
84.95
94.14
95.78
98.37
-
-
-
-
GeoDTR †
86.21
95.44
96.72
98.77
64.52
88.59
91.96
98.74
PaSS-KD †
87.20
94.30
95.67
97.85
66.81
88.03
90.87
98.02
Sample4Geo
90.81
96.74
97.48
98.77
71.51
92.42
94.45
98.70
Table 2: Performance comparison on CVACT dataset. † indicates applying polar transform to aerial images. Best results in bold, second-best in underlined. “-” means the score is not provided in the original paper.
Set
Methods
FoV= 360∘
FoV= 180∘
FoV= 90∘
FoV= 70∘
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
CVUSA
CVM
16.25
38.86
49.41
88.11
7.38
22.51
32.63
75.38
2.76
10.11
16.74
55.49
2.62
9.30
15.06
21.77
CVFT
23.38
44.42
55.20
86.64
8.10
24.25
34.47
75.15
4.80
14.84
23.18
61.23
3.79
12.44
19.33
55.56
DSM
78.11
89.46
92.90
98.50
48.53
68.47
75.63
93.02
16.19
31.44
39.85
71.13
8.78
19.90
27.30
61.20
SEH
85.37
93.51
95.81
-
53.73
72.32
78.96
-
16.58
32.17
40.34
-
7.82
18.84
25.59
-
L2LTR
-
-
-
-
56.69
80.86
87.75
98.01
26.92
50.49
60.41
86.88
13.95
33.07
43.86
77.65
Table 3: Comparison of images with unknown orientation and different FoVs on the CVACT and CVUSA datasets. ‘’-” means the score is not provided in the original paper. The best performance is in bold and the second-best performance is underline. Note that the model is trained and tested on FoV-specific images.
Methods
CVACT_val
CVACT_test
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
Baseline
85.37
94.83
96.23
98.50
66.50
87.94
90.08
98.49
Baseline + SCM
90.21
96.07
96.99
98.47
71.90
91.71
93.73
98.36
Baseline + HFIG
89.80
95.69
96.67
98.30
71.72
91.68
93.46
98.32
Baseline + SCM + HFIG (Ours)
91.02
96.83
97.53
98.86
72.28
92.69
94.47
98.69
Table 4: Ablation study on CVACT dataset. We evaluate the Effectiveness of each component in our method. Best results are in bold.
strategie
CVACT_val
CVACT_test
R@1
R@5
R@10
R@1%
R@1
R@5
R@10
R@1%
Repeated Scan
90.14
96.38
97.29
98.66
71.01
91.68
93.68
98.40
Bidirectional Scan
90.29
95.94
96.92
98.55
71.34
91.86
93.95
98.45
Cross Scan
90.38
96.06
97.01
98.45
71.84
91.69
93.66
98.36
Rotation Scan (Ours)
91.02
96.83
97.53
98.86
72.28
92.69
94.47
98.69
Table 5: Effectiveness of scanning strategies in SCM. Best results are in bold.
Figure 3: Heatmap visualization of the Baseline and SLGeo model. Best viewed on screen with zoom-in.
Robust cross-view geo-localization (CVGL) remains challenging despite the surge in recent progress. Existing methods still rely on field-of-view (FoV)-specific training paradigms, where models are optimized under a fixed FoV but collapse when tested on unseen FoVs and unknown orientations. This limitation necessitates deploying multiple models to cover diverse variations. Although studies have explored dynamic FoV training by simply randomizing FoVs, they failed to achieve robustness across diverse conditions -- implicitly assuming all FoVs are equally difficult. To address this gap, we present SinGeo, a simple yet powerful framework that enables a single model to realize robust cross-view geo-localization without additional modules or explicit transformations. SinGeo employs a dual discriminative learning architecture that enhances intra-view discriminability within both ground and satellite branches, and is the first to introduce a curriculum learning strategy to achieve robust CVGL. Extensive evaluations on four benchmark datasets reveal that SinGeo sets state-of-the-art (SOTA) results under diverse conditions, and notably outperforms methods specifically trained for extreme FoVs. Beyond superior performance, SinGeo also exhibits cross-architecture transferability. Furthermore, we propose a consistency evaluation method to quantitatively assess model stability under varying views, providing an explainable perspective for understanding and advancing robustness in future CVGL research. Codes will be available upon acceptance.
Yang Chen, Xieyuanli Chen, Junxiang Li +2
College of Intelligence Science and Technology, National University of Defense Technology
Cross-view geo-localization is challenging due to drastic viewpoint changes and large appearance discrepancies between street-level and satellite imagery. Although existing methods often use geometric warping to expose co-visible cues, such transformations rely on restrictive spatial assumptions and inevitably introduce severe visual distortions under view-dependent visibility, yielding noisy supervision and fragile correspondences. To overcome this, we propose a novel joint-view consensus-guided learning framework that entirely bypasses explicit geometric warping. Instead of forcing rigid spatial alignment, we dynamically mine and adaptively strengthen a semantic consensus directly within the feature space. Specifically, an auxiliary joint-view pathway during training enables direct cross-view interaction, allowing each view to selectively aggregate corroborative evidence into a unified consensus representation. To resolve feature heterogeneity among the single- and joint-view streams, we introduce global pattern probes acting as a semantic dictionary to project divergent modalities into a strictly aligned metric space. Guided by a consensus-mediated contrastive objective, single-view embeddings are explicitly pulled toward the joint-view anchor during training, distilling this consensus-mining capability into the single-view encoders for robust retrieval at inference. Extensive experiments demonstrate that our method achieves state-of-the-art performance across four standard benchmarks, underscoring the importance of discovering cross-view semantic consensus for reliable geo-localization.
Zhuo Song, Lian Xu, Runqing Jiang +4
Sun Yat-sen University, Shenzhen, China · The University of Western Australia, Perth, Australia
Generalizable cross-view geo-localization aims to match the same location across views in unseen regions and conditions without GPS supervision. Its core difficulty lies in severe semantic inconsistency caused by viewpoint variation and poor generalization under domain shift. Existing methods mainly rely on 2D correspondence, but they are easily distracted by redundant shared information across views, leading to less transferable representations. To address this, we propose GeoLink, a 3D-aware semantic-consistent framework for Generalizable cross-view geo-localization. Specifically, we offline reconstruct scene point clouds from multi-view drone images using VGGT, providing stable structural priors. Based on these 3D anchors, we improve 2D representation learning in two complementary ways. A Geometric-aware Semantic Refinement module mitigates potentially redundant and view-biased dependencies in 2D features under 3D guidance. In addition, a Unified View Relation Distillation module transfers 3D structural relations to 2D features, improving cross-view alignment while preserving a 2D-only inference pipeline. Extensive experiments on multiple benchmarks show that GeoLink consistently outperforms state-of-the-art methods and achieves superior generalization across unseen domains and diverse weather environments.