Synergistic spatial intelligence between UAVs and satellites is indispensable for emergency response and security operations, as it uniquely integrates macro-scale global coverage with dynamic, real-time local perception. However, the capacity of Vision-Language Models (VLMs) to master this complex interplay remains largely unexplored. This gap persists primarily because existing benchmarks are confined to isolated Unmanned Aerial Vehicle (UAV) videos or static satellite imagery, failing to evaluate the dynamic local-to-global spatial mapping essential for comprehensive cross-view reasoning. To bridge this gap, we introduce LinkS2Bench, the first comprehensive benchmark designed to evaluate VLMs' wide-area, dynamic cross-view spatial intelligence. LinkS2Bench links 1,022 minutes of dynamic UAV footage with high-resolution satellite imagery covering over 200 km2. Through an LMM-assisted pipeline and rigorous human annotation, we constructed 17.9k high-quality question-answer pairs comprising 12 fine-grained tasks across four dimensions: perception, localization, relation, and reasoning. Evaluations of 18 representative VLMs reveal a substantial gap compared to human baselines, identifying accurate cross-view dynamic alignment as the critical bottleneck. To alleviate this, we design a Cross-View Alignment Adapter, demonstrating that explicit alignment significantly improves model performance. Furthermore, fine-tuning experiments underscore the potential of LinkS2Bench in advancing VLM adaptation for complex spatial reasoning.
Figures & tables
Figure 1. Overview of LinkS 2 Bench. (a) Compared with existing benchmarks, LinkS 2 Bench uniquely combines wide-area coverage, UAV-satellite cross-view reasoning, and dynamic spatiotemporal understanding. (b) Evaluation of representative VLMs against human performance across 12 fine-grained tasks reveals a substantial gap in cross-view spatial intelligence.
Benchmark
View Setting
Input Modality
Dynamic
Cross-view
# Tasks
# QA Pairs
VSI-Bench (CVPR 2025) ( Yang et al., 2025a )
Ground
Video
✓
×
8
5.0k
All-Angles-Bench (AAAI 2026) ( Yeh et al., 2025 )
Ground
Image
×
×
6
2.1k
Open3D-VQA (ACM MM 2025) ( Zhang et al., 2025c )
UAV
Image
×
×
7
73.3k
AirCopBench (AAAI 2026) ( Zha et al., 2025 )
UAV
Image
×
×
14
14.6k
SpatialSky-Bench (CVPR 2026) ( Zhang et al., 2025a )
UAV
Image
×
×
13
1.0k
UrbanVideo-Bench (ACL 2025) ( Zhao et al., 2025a )
UAV
Video
✓
×
16
5.2k
Table 1. Comparison of LinkS 2 Bench with representative spatial intelligence benchmarks. Comparison of benchmarks by view setting, input modality, cross-view ability, dynamic scenes, and number of QA pairs.
Figure 2. Benchmark Statistics. (a) The distribution of tasks across four main categories. (b) Distribution of data sources across different video durations.
Figure 3. LinkS 2 Bench comprises 12 fine-grained task types categorized into four primary evaluation dimensions: Perception, Location, Relation, and Reasoning. For visualization purposes, the satellite images are cropped to highlight the annotations; the actual benchmark samples preserve the full original spatial context.
Figure 4. LinkS 2 Bench curation pipeline includes data collection, question generation, human annotation and quality control.
Method
Rank
Avg.
Perception
Localization
Relation
Reasoning
TV
SA
ZC
Avg.
EG
TM
TG
Avg.
RN
RD
OA
Avg.
SO
RI
RR
Avg.
Baseline
Random
-
50.0
25.0
-
-
25.0
25.0
-
-
12.5
20.0
25.0
19.2
16.7
33.3
50.0
33.3
LinkS 2 Bench (tiny) Perf.
Human Level
-
91.3
98.3
96.7
85.8
93.6
90.0
91.7
88.3
90.0
93.3
95.0
86.7
91.7
85.0
93.3
91.2
89.8
Gemini-3.1-Pro
-
52.7
76.7
60.0
60.7
65.8
53.3
48.3
41.7
47.8
53.3
28.3
48.3
43.3
50.0
46.7
65.0
53.9
Table 2. Results (%) on LinkS 2 Bench for existing various VLMs on 12 task types across 4 evaluation dimensions. The former section shows existing popular models’ results. The latter section demonstrates model adaption experimental results.
Figure 5. Pairwise task correlations based on VLM performance. Higher values suggest that VLMs perform more similarly on the two tasks, indicating a greater overlap in the abilities required to solve them.
Figure 6. Distribution of error types via manual inspection.
Figure 7. Overview of Cross-View Alignment Adapter and Alignment-Guided VLM Reasoning.
Humans can effortlessly reason about scenes across different viewpoints, yet it remains unclear whether Vision-Language Models (VLMs) possess similar cross-view spatial abilities. Satellite-street scene pairs, with their complex contexts and extreme viewpoint variations, provide an ideal testbed. Motivated by this, we introduce CVSBench, a large-scale benchmark for evaluating cross-view spatial reasoning through satellite-street pairs. This benchmark supports multiple tasks, including cross-view VQA, cross-view grounding, and viewpoint identification. CVSBench comprises 3,297 cross-view image groups with 9,468 object-level annotations and 40,679 question-answer (QA) pairs, enabling systematic and controlled evaluation of cross-view spatial reasoning. Extensive evaluations reveal that advanced VLMs struggle to maintain object-level and layout consistency under drastic viewpoint changes. To bridge this gap towards human-like spatial cognition, we investigate two categories of approaches: spatially grounded reasoning and the incorporation of cognitive map inputs. Our findings demonstrate that language-only reasoning yields marginal improvements, while incorporating visual spatial imagination via a 3D scene imagination pipeline substantially improves cross-view reasoning. These results highlight the necessity of explicit visual-spatial representations for robust spatial cognition in VLMs. Our data and code are released at https://huggingface.co/datasets/zlyzlyzly/CVSBench.
Ruixun Liu, Lingyu Zhang, Lanxuan Xue +3
Faculty of Electronic and Information Engineering, Xi’an Jiaotong University · Ministry of Education Key Laboratory of Intelligent Networks and Network Security
Spatial intelligence is fundamental to embodied agents, yet existing benchmarks focus on local spatial perception from single or few viewpoints, overlooking global spatial awareness over continuous, long-horizon visual streams. To address this limitation, we introduce the Global-Spatial-Temporal Benchmark (GST-Bench), a VQA benchmark for global spatial intelligence in video understanding, comprising human-verified questions derived from 6,790 minutes of synthetically generated video. It requires models to perform accurate spatial inference from novel viewpoints unseen in the input video and to map egocentric observations onto global top-down images. A comprehensive evaluation of 22 state-of-the-art VLMs exposes a striking gap between models and humans: the strongest zero-shot model attains only 42.68, far below the human score of 79.08. To probe the cause of this gap, we construct GST-Bench-Local and find that models, despite strong local spatial understanding under the same task formulation, still fail to consolidate long-horizon observations into a globally consistent scene representation. We further provide GST-Train, a dataset for global spatial reasoning, as a complementary resource to facilitate future research on this challenge.
Qifeng Zhang, Kaixiang Huang, Heng Dong +6
ByteDance Seed · Zhejiang University · National University of Singapore +1
We push the frontier of large-scale spatial intelligence in Vision-Language Models (VLMs) and introduce the first benchmark that probes geographical layout understanding from real-world videos, spanning up to 1km distances. Inspired by the cognitive science literature, we evaluate models against the hierarchical stages of human spatial awareness: anchoring via landmarks, connecting them through routes, and integrating these into global mental maps. Extensive experiments reveal a fundamental divergence in how current AI models process spatial information. Instead of utilising true path integration or forming geometric survey knowledge, we find that VLMs rely almost entirely on 2D visual recognition and text-matching to bypass complex spatial reasoning. The benchmark is publicly available at https://perception-test-challenge.github.io/kilometervision.html.
Aravindh Mahendran, Michael King, Matthew Koichi Grimes +12
Google DeepMind, Berlin, Germany · Google DeepMind, London, UK · Princeton University, Princeton, USA +3