Organizations: College of Computer Science, Beijing University of Technology · Zhengzhou University · Zhejiang University · Beijing Institute of Technology · Beijing Electronic Science and Technology Institute · Intelligent Science & Technology Academy of CASIC
Cross-view object geo-localization (CVOGL) locates a target in satellite imagery using drone or street-view queries. Existing methods train separate detectors for each viewpoint, leading to parameter redundancy and impeding cross-view knowledge sharing. Moreover, top-ranked satellite candidates are often visually similar, so visual appearance and categorical labels alone are insufficient to resolve such ambiguity. To address these, we propose MVLGeo, an efficient framework designed to unify multiple viewpoints and reduce model redundancy. First, we introduce environmental contextual text from the query view as cues to distinguish visually similar candidates via Vision-Language Reranking (VL-Rerank). Second, we design a multi-view Mixture-of-Experts architecture (MV-MoE) with a shared encoder and view-specific experts to reduce redundancy and promote knowledge sharing, while cross-view contrastive learning aligns their representations for consistency. Third, we introduce an adaptive elliptical prior (ESAM-Prior) as auxiliary positional encoding for anisotropic geometric perception. Extensive experiments on the CVOGL benchmarks confirm that MVLGeo, as a unified model for multiple query viewpoints, achieves state-of-the-art performance, demonstrating robustness to input degradation and generalization across viewpoints. Code and models will be available on GitHub to facilitate future work.
Figures & tables
Figure 1: Illustration of the VL-Rerank framework. A VLM generates environmental context (e.g., road topology) from query images. These descriptions rerank satellite candidates, filtering out visually similar but contextually mismatched bounding boxes and identifying the correct target.
Figure 2: Comparison of the proposed MV-MoE and independent models. MV-MoE shares one encoder and two lightweight experts (UAV, SVI) for four input variants (UAV Vis, UAV Gray, SVI Vis, SVI Gray), reducing parameters by 38% while maintaining Acc@50 (%).
Figure 3: Schematic diagram of the MVLGeo network architecture. (A) Input-guided features are generated using SAM and shape priors. (B) Balancing heterogeneous input channels. (C–D) Using a Mixture-of-Experts (MoE) architecture with hard gating to perform cross-view feature extraction, fusion, and detection from UAV and SVI perspectives. (E) Aligning cross-view features during the training phase via InfoNCE contrastive learning. (F) Performing multi-image comparison and reranking by combining environmental semantic descriptions and image cropping to output the Top-1 matching result.
Figure 4: Recall@K reveals that the ground-truth box often appears among top candidates but is not ranked first.
Figure 5: Shape prior generation pipeline. (a) Query image, (b) SAM mask and fitted ellipse, (c) Anisotropic Gaussian heatmap.
Ground → Satellite
Drone → Satellite
Method
Validation
Test
Validation
Test
Models
Acc@25%
Acc@50%
Acc@25%
Acc@50%
Acc@25%
Acc@50%
Acc@25%
Acc@50%
CVM-Net [ 11 ]
5.09
0.87
4.73
0.51
20.04
3.47
20.14
3.29
2
RK-Net [ 18 ]
8.67
0.98
7.40
0.82
19.94
3.03
19.22
2.67
2
L2LTR [ 35 ]
12.24
1.84
10.69
2.16
38.68
5.96
38.95
6.27
2
Polar-SAFA [ 26 ]
19.18
2.71
20.66
3.19
36.19
6.39
37.41
6.58
2
Table 1: Comparison of Acc@25/50 (%) and number of models on the CVOGL dataset. Bold denotes the best results.
Method
Params
Models
Total
DetGeo
73.8M
2
147.6M
OCGNet
74.8M
2
149.6M
MVLGeo (Ours)
91.0M
1
91.0M
Table 2: Comparison of model parameters and deployment costs between existing dual-detector methods and our single unified model.
VL- Rerank
MV- MoE
ESAM- Prior
UAV Acc@50 (%)
SVI Acc@50 (%)
Params
×
×
×
57.66
42.24
73.8M × 2
×
×
✓
62.58
46.76
73.8M × 2
×
✓
×
52.14
41.35
91.0M × 1
×
✓
✓
56.63
44.09
91.0M × 1
✓
✓
✓
66.04
47.24
91.0M × 1
Table 3: Ablation study of each component in MVLGeo on the UAV and SVI test sets.
Figure 6: Contributions of each component to the UAV and SVI test sets, and corresponding parameter relationships.
#
CL λ
BN γ init
Val Best
Test Best
1
0.1
1.0 (identity)
49.29
50.36
2
0.1
Random
44.74
43.91
3
0.5
1.0
47.52
46.87
4
0.0
1.0
49.59
49.23
Table 4: Average Acc@50 (%) across UAV and SVI on validation and test sets under MV-MoE configurations.
Figure 7: Cross-dataset localization accuracy distribution on UAV and SVI scenes. The yellow value line denotes the Pareto trade-off boundary, and points above the line achieve superior overall performance.
Figure 8: ESAM-Prior ablation results on UAV and SVI datasets. (a) Validation Acc@50 (%) over training epochs. (b) Acc@50 (%) across object aspect ratio buckets partitioned into Low, Medium and High groups.
Test modality
Single baseline
MV-MoE unified
Retention
UAV Vis (UV)
62.6
56.6
90.4%
UAV Gray (UG)
62.4
54.5
87.3%
SVI Vis (SV)
46.8
44.0
94.0%
SVI Gray (SG)
45.9
44.1
96.1%
4-View Avg
—
49.8
—
Table 5: Unified MV-MoE generalizes across four modality-viewpoint combinations with minimal accuracy degradation under grayscale inputs.
Cross-View Object Geo-Localization (CVOGL) aims to locate an object of interest in a query image within a corresponding satellite image. Existing methods typically assume that the query image contains only a single object, which does not align with the complex, multi-object geo-localization requirements in real-world applications, making them unsuitable for practical scenarios. To bridge the gap between the realistic setting and existing task, we propose a new task, called Cross-View Multi-Object Geo-Localization (CVMOGL). To advance the CVMOGL task, we first construct a benchmark, CMLocation, which includes two datasets: CMLocation-V1 and CMLocation-V2. Furthermore, we propose a novel cross-view multi-object geo-localization method, MOGeo, and benchmark it against existing state-of-the-art methods. Extensive experiments are conducted under various application scenarios to validate the effectiveness of our method. The results demonstrate that cross-view object geo-localization in the more realistic setting remains a challenging problem, encouraging further research in this area.
Bo Lv, Qingwang Zhang, Le Wu +2
College of Computer Science and Software Engineering, Shenzhen University
Cross-view object geo-localization (CVOGL) aims to locate a target object from a query view (e.g., ground or drone) within a geo-tagged reference image (e.g., satellite). Existing approaches heavily rely on 2D appearance matching and are constrained by limited datasets lacking geometric metadata, diverse prompts, and standard field-of-view imagery. To address these intertwined challenges, we first introduce \dataset, a large-scale, high-fidelity building dataset comprising over 220,000 ground-satellite and drone-satellite pairs. It provides multi-modal prompts (points, boxes, masks) and camera poses to enable flexible target referring and explicit spatial modeling. Furthermore, we propose a novel single-stage Geometry-Aware Geo-localization framework (GAGeo), built upon the permutation-equivariant 3D foundation model π3. By seamlessly integrating visual features, referring prompts, and learnable task tokens, our model adapts the inherited 3D prior to jointly predict bounding boxes, segmentation masks, and camera poses in a single forward pass. Additionally, we introduce a contrastive loss that utilizes the satellite view as a universal anchor, implicitly aligning ground and drone representations to enable zero-shot ground-to-drone localization without requiring triplet training data. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods, exhibiting exceptional generalization ability in unseen scenes and novel cross-view setups.
Liyao Wang, Ruipu Wu, Haojun Xu +3
Beihang University, Beijing, China · Meituan, Beijing, China
Cross-view geo-localization (CVGL) aims to estimate geographic locations by matching images captured from different viewpoints, such as drone and satellite views. Existing methods mainly rely on visual representations, but often fail to jointly model fine-grained structural correspondences and semantic priors, making them prone to confusion between visually similar but semantically different regions, and thus limiting robustness under large viewpoint variations. To address these challenges, we propose \textbf{S3Geo}, a structure-semantic synergistic learning framework for cross-view matching. Specifically, we first introduce a Decoupled Query Pooling (DQP) module to extract a compact set of region-aware features from dense tokens, enabling explicit modeling of local structural patterns. We then design a query-level contrastive learning scheme with an optimal transport (OT)-based formulation to establish soft correspondences under cross-view spatial misalignment. Furthermore, we incorporate a Semantic Knowledge Distillation (SKD) strategy from a frozen CLIP teacher to transfer semantic priors and relational structures, thereby improving discrimination on hard negatives. By operating synergistically, the semantic priors provide robust contextual filtering, which guides the structural module to establish precise spatial alignments. Experiments on the University-1652 and SUES-200 datasets demonstrate that \textbf{S3Geo} consistently outperforms state-of-the-art approaches without increasing inference complexity, validating the effectiveness of jointly modeling structural and semantic information for CVGL.
Ziqian Mo, Hill Zhang, Haosheng Tan +2
The Hong Kong University of Science and Technology (Guangzhou) Guangzhou, China · Claremont McKenna College Claremont, California, USA