AirZoo: A Unified Large-Scale Dataset for Grounding Aerial Geometric 3D Vision
Paper • 2604.26567 • Published • 2
Datasets:
Request access to AirZoo Data
For non-commercial research or education. Access is granted automatically after you agree.
Access is for non-commercial research or education only. Do not redistribute the dataset. Follow Cesium and related provider terms for map-derived content.
Log in or Sign Up to review the conditions and access this dataset content.
AirZoo Data
Synthetic aerial training renders from the AirZoo project (RGB, metric depth, and poses).
This repo is gated. Request access on this page before downloading. Approval is automatic after you agree.
Renders were produced with Cesium for Unreal, Unreal Engine, and AirSim. Their licenses and terms still apply. Access here does not allow redistribution of the data.
Organize RAW data
Each pose file lists one line per frame: frame_id lon lat alt roll pitch yaw.
Camera is pinhole with no distortion. All sequences share the same (K) at a given RGB size; pick the row that matches your images.
| Resolution | (f_x, f_y) | (c_x, c_y) | FOV (H × V) |
|---|---|---|---|
| 1600×1200 | 1931.7, 1931.7 | 800, 600 | 45.0° × 34.5° |
| 1920×1080 | 2317.6, 2317.6 | 960, 540 | 45.0° × 26.2° |
Contents
AirZoo_TrainingData/— sequence archives
Related
- Engine: https://huggingface.co/datasets/choyaa/AirZoo-Engine
- Code: https://github.com/nudt-sawlab/AirZoo
- Benchmarks: https://huggingface.co/datasets/RingoWRW97/AirZoo-Real
- Website: https://nudt-sawlab.github.io/AirZoo/
Citation
@article{cheng2026airzoo,
title={AirZoo: A Unified Large-Scale Dataset for Grounding Aerial Geometric 3D Vision},
author={Cheng, Xiaoya and Wu, Rouwan and Liu, Xinyi and Cui, Zeyu and Liu, Yan and Zhao, Na and Liu, Yu and Zhang, Maojun and Yan, Shen},
journal={arXiv preprint arXiv:2604.26567},
year={2026}
}
- Downloads last month
- 405