AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground Video-based Person Re-Identification
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.
- 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
We propose AG-VPReID-Net, a three-stream architecture specifically designed for aerial-ground person ReID:
- Adapted Temporal-Spatial Stream addressing motion pattern inconsistencies and temporal feature learning
- Normalized Appearance Stream using physics-informed techniques to tackle resolution and appearance changes
- Multi-Scale Attention Stream handling scale variations across drone altitudes
| Method | MARS (mAP/R-1) |
LS-VID (mAP/R-1) |
iLIDS-VID (R-1) |
G2A-VReID (mAP/R-1) |
AG-VPReID (A2G) (mAP/R-1) |
AG-VPReID (G2A) (mAP/R-1) |
|---|---|---|---|---|---|---|
| LSTRL | 86.8/91.6 | 82.4/89.8 | 92.2 | - | 61.7/71.3 | 56.7/66.5 |
| CLIP-ReID | 88.1/91.7 | 80.6/88.8 | - | - | 62.3/71.6 | 57.2/66.8 |
| AG-VPReID-Net | 91.5/93.2 | 87.3/93.2 | 96.3 | 81.3/73.1 | 64.0/71.9 | 58.0/75.6 |
| Stream | 15m (A2G/G2A) |
30m (A2G/G2A) |
80m (A2G/G2A) |
120m (A2G/G2A) |
|---|---|---|---|---|
| Stream 1 | 80.28/83.25 | 78.76/83.03 | 67.24/67.07 | 52.47/62.31 |
| Stream 2 | 69.75/72.87 | 68.13/71.41 | 52.62/52.22 | 38.12/45.43 |
| Stream 3 | 74.25/77.18 | 74.01/78.91 | 52.53/56.59 | 35.13/52.56 |
| Combined | 80.66/83.92 | 79.00/83.66 | 67.63/67.82 | 53.00/63.32 |
The AG-VPReID dataset is available at Download Link.
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}
}