README.md

metadata

license: mit
task_categories:
  - image-text-to-text
  - video-text-to-text
tags:
  - uav
  - spatial-intelligence
  - vqa

SpatialUAV: Benchmarking Spatial Intelligence for Low-Altitude UAV Perception, Collaboration, and Motion

Paper | GitHub

SpatialUAV is a benchmark for evaluating spatial intelligence in real low-altitude UAV scenarios. It covers perception, spatial relation reasoning, aerial-aerial collaboration, aerial-ground collaboration, and UAV motion understanding under a unified visual-question-answer format.

Highlights

• 4,331 curated instances from real low-altitude UAV images, videos, and metadata.
• 14 task types across semantic discrimination, spatial relations, aerial-aerial collaboration, aerial-ground collaboration, and motion understanding.
• 7 visual input configurations covering single images, paired views, candidate-view selection, annotated images, and ordered video frames.
• 9 answer formats, including option labels, region IDs, region pairs, bounding boxes, angle-distance values, movement directions, and free-form text.
• Task-specific evaluation for heterogeneous outputs instead of relying on one generic text metric.

Benchmark Overview

Group	Instances	Task Types	Main Capability
Semantic Discrimination	599	Region Recognition, Anomaly Detection	Recognize queried objects and safety-critical regions
Spatial Relation	716	Direction Recognition, Distance Comparison	Infer direction and relative depth from UAV views
Aerial-Aerial Collaboration	1,231	Collaboration Recognition, Shared Association, Object Matching, Camera Transformation, Occlusion Removal	Match and reason across multiple UAV viewpoints
Aerial-Ground Collaboration	785	Shared Association, Collaboration Recognition, View Translation, Path Planning	Align aerial and ground observations
Motion Understanding	1,000	Global Motion	Describe UAV/camera motion over ordered frames

Data Format

The annotation file is JSONL. Each line is one benchmark instance:

{
  "id": "Region_Recognition_00001",
  "image": ["./SpatialUAV/samples_Single_Image/img0001.jpg"],
  "conversations": [
    {
      "from": "human",
      "value": "Which regions in the image contain a parking lot? Answer with only the region labels, formatted exactly like `Region 1, 2`. No explanation."
    }
  ],
  "source": "SpatialUAV",
  "GT": "Region 3, 4."
}

Expected dataset layout after download:

SpatialUAV/
  annotations.jsonl
  annotations_subset_20pct_per_task.jsonl
  samples_Single_Image/
  samples_A2A_Pured/
  samples_A2A_detected/
  samples_A2A_Occlusion_Removal/
  samples_A2G_Pured/
  samples_A2G_detected/
  samples_A2G_Path_Planning/
  samples_Motion_Understanding_Frames/

Use annotations_subset_20pct_per_task.jsonl for quick checks and annotations.jsonl for full benchmark evaluation.

Citation

@article{zhang2026spatialuav,
  title   = {SpatialUAV: Benchmarking Spatial Intelligence for Low-Altitude UAV Perception, Collaboration, and Motion},
  author  = {Zhang, Haoyu and Liu, Meng and Xiang, Qianlong and Wang, Kun and Wang, Yaowei and Nie, Liqiang},
  journal = {arXiv preprint arXiv:2606.27876},
  year    = {2026}
}