🔥【AAAI 2026】Text-based Aerial-Ground Person Retrieval

News • Setup • Dataset • Config • Training • Citation

---

📢 News

• [2025-12] Code and dataset released
• [2025-11] Paper uploaded to arXiv

---

🛠️ Setup

Requirements

• Hardware: 4 × NVIDIA RTX 3090 GPUs (24GB VRAM recommended)
• Software: Python 3.8+ (Recommended), CUDA 11.x+

Installation

Clone the repository and install dependencies:

git clone https://github.com/Flame-Chasers/TAG-PR.git
cd TAG-PR
pip install -r requirements.txt

---

📂 Dataset Preparation

We provide the processed dataset via Quark Cloud Drive.

1. Download: Click here to download (Access Code: 8pE6)
2. Organize: Extract and arrange the files as follows:

dataset/
├── anno_dir/
│   ├── train_reid.json
│   └── test_reid.json
└── images/
    ├── 0001.jpg
    ├── 0002.jpg
    └── ...

---

⚙️ Configuration

Modify the configuration file located at config/s.config.yaml.

# Data Paths
anno_dir: "/path/to/dataset/anno_dir"      # ⚠️ Absolute path to annotation JSONs
image_dir: "/path/to/dataset/images"       # ⚠️ Absolute path to image directory

# Model Settings
model:
  checkpoint: "/path/to/clip/ViT-B-16.pt"  # Path to pre-trained CLIP weights
  # ... other model params

---

🚀 Training

We support multi-GPU training and shell script execution.

Option 1: Torchrun

Use torchrun for distributed data parallel (DDP) training:

CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun \
  --rdzv_id=12345 \
  --rdzv_backend=c10d \
  --rdzv_endpoint=localhost:0 \
  --nnodes=1 \
  --nproc_per_node=4 \
  main.py

Option 2: Shell Script

You can also use the provided shell script wrapper:

bash shell/train.sh

---

📝 Citation

If you find this project useful for your research, please consider citing our paper:

@article{zhou2025text,
  title={Text-based Aerial-Ground Person Retrieval},
  author={Zhou, Xinyu and Wu, Yu and Ma, Jiayao and Wang, Wenhao and Cao, Min and Ye, Mang},
  journal={arXiv preprint arXiv:2511.08369},
  year={2025}
}

📄 License

This project is released under the MIT License.