DroneKey++
[Project] [Paper]
6DroneSyn Dataset
A large-scale synthetic benchmark for drone 3D pose estimation — 52,920 high-resolution images across 7 drone models and 88 outdoor backgrounds.
Why 6DroneSyn?
Accurate 3D pose estimation of drones is essential for security and surveillance systems. However, existing datasets are small-scale, limited to single drone models, and collected under constrained environments, making reliable validation of generalization difficult. Since obtaining precise 3D drone pose annotations in real-world settings is nearly impossible due to safety regulations, equipment limitations, and cost constraints, synthetic data is the most practical path forward.
6DroneSyn is a comprehensive synthetic benchmark designed to overcome these limitations. It was generated using 360-degree panoramic synthesis with real-world background images, effectively minimizing the domain gap between synthetic and real data. Feature distribution analysis using PCA and t-SNE confirms that our synthetic data closely overlaps with real drone imagery, achieving both realism and diversity.
7 DJI Drone Types
The dataset covers a wide range from small consumer to professional-grade drones, enabling generalization across diverse drone types.
Comprehensive Annotation Types
Each frame provides rich, automatically generated annotations for training and evaluation of various drone perception tasks.
Dataset Structure
13 synthetic scenes featuring diverse drone motion patterns and 88 background variations from 22 real-world environments. Each scene generates independent sequences for each drone model.
Download 6DroneSyn
The dataset is publicly available for research purposes. Please cite our paper if you use this dataset in your work. 🙏
↓ Coming soon
Cite Our Work
@inproceedings{hwang2026dronekey++,
title = {DroneKey++: A Size Prior-free Method and New Benchmark
for Drone 3D Pose Estimation from Sequential Images},
author = {Hwang, Seo-Bin and Cho, Yeong-Jun},
booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
year = {2026}
}
Data Samples
⚠️ Note: Large files — loading may be slow.
Scene # 01
Drone Type: 01
120 Frames
Scene # 02
Drone Type: 02
120 Frames
Scene # 03
Drone Type: 03
120 Frames
Scene # 04
Drone Type: 04
120 Frames
Scene # 11
Drone Type: 05
360 Frames
Scene # 12
Drone Type: 06
360 Frames
Scene # 13
Drone Type: 07
360 Frames