AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground Video-based Person Re-Identification

Huy Nguyen¹, Kien Nguyen¹, Akila Pemasiri¹, Feng Liu², Sridha Sridharan¹, Clinton Fookes¹

¹School of Electrical Engineering and Robotics, Queensland University of Technology  ²Department of Electrical and Computer Engineering, Drexel University

CVPR 2025

AG-VPReID is a comprehensive benchmark dataset for aerial-ground video-based person re-identification, offering unprecedented scale and diversity with footage from drones (15-120m altitude), CCTV cameras, and wearable devices.

arXiv | paper | dataset

Key Features

• The largest aerial-ground video dataset with 6,632 identities, 32,321 tracklets, and 9.6 million frames
• Multiple platforms combining aerial drones (at various altitudes: 15m, 30m, 80m, 120m), CCTV cameras, and wearable cameras
• Rich annotations including clothing changes and 15 soft-biometric attributes
• Real-world challenges including extreme viewpoint variations, scale differences, and occlusions
• Diverse environments with outdoor campus scenes captured over 20 days

AG-VPReID-Net Framework

We propose AG-VPReID-Net, a three-stream architecture specifically designed for aerial-ground person ReID:

1. Adapted Temporal-Spatial Stream addressing motion pattern inconsistencies and temporal feature learning
2. Normalized Appearance Stream using physics-informed techniques to tackle resolution and appearance changes
3. Multi-Scale Attention Stream handling scale variations across drone altitudes

Performance

Comparison with State-of-the-Art Methods

Performance Across Drone Altitudes

Dataset Access

The AG-VPReID dataset is available at Download Link.

Citation

If you use our dataset or code in your research, please cite our paper:

@inproceedings{nguyen2025agvpreid,
  title={AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground Video-based Person Re-Identification},
  author={Nguyen, Huy and Nguyen, Kien and Pemasiri, Akila and Liu, Feng and Sridharan, Sridha and Fookes, Clinton},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2025}
}