AG-VPReID: Aerial-Ground Video-based Person Re-Identification

Official repository for the AG-VPReID Competition

Dataset Access

The dataset is available for download on Kaggle. If you experience download issues, please check the GitHub issue comment for troubleshooting steps.

Data Structure

The dataset follows a hierarchical organization: {ID}/{Tracklets}/{Frames}

• Person ID: Unique identifier for each person
• Camera Types:
  • C0/C1: CCTV cameras (ground-view)
  • C2/C3: Wearable cameras (ground-view)
  • C4/C5: UAV cameras (aerial-view)
• Frame Format: Fxxx (where xxx is the frame number)

Evaluation Protocol

The competition evaluates performance on two distinct scenarios:

1. Aerial-to-Ground: Queries from aerial cameras, gallery from ground cameras
2. Ground-to-Aerial: Queries from ground cameras, gallery from aerial cameras

Benchmark Results

Method	Aerial-Ground	Ground-Aerial	Overall
	R1/R5/R10/mAP	R1/R5/R10/mAP	R1/R5/R10/mAP
1st Team	-/-/-/-	-/-/-/-	-/-/-/-
2nd Team	-/-/-/-	-/-/-/-	-/-/-/-
3rd Team	-/-/-/-	-/-/-/-	-/-/-/-
baseline_tfclip	0.6308/0.7516/0.7989/0.6552	0.6449/0.7986/0.8397/0.6707	0.6375/0.7740/0.8183/0.6626

Setup Instructions

1. Clone this repository.
2. Download the AG-VPReID dataset.
3. Organize the dataset as follows:

datasets/
    AG-VPReID/
      train/
      case1_aerial_to_ground/
        gallery/
        query/
      case2_ground_to_aerial/
        gallery/
        query/
      attributes/

4. For quick start, follow instruction in baseline

Submission Guidelines

1. Generate prediction files for both test cases.
2. Merge submission_case1_aerial_to_ground.csv and submission_case2_ground_to_aerial.csv maintaining the header.
3. Submit the merged file to the AG-VPReID competition on Kaggle for evaluation.

FAQs

This section will be updated regularly as new information becomes available.

📖 Citation

If you find our work useful, please consider citing our paper:

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