Home

Awesome

VDT-AGPReID

View-decoupled Transformer for Person Re-identification under Aerial-ground Camera Network (CVPR'24) [paper_link]

[1] Dataset: CARGO

CARGO

Introduction

Setting

Annotation

Annotations are preserved in the name of each image by the format ``camID_time_personID_index.jpg''.

For example, ``Cam2_day_2519_320.jpg'' means that:

License

[2] Method: View-decoupled Transformer

VDT

Requirements

Step1: Prepare enviorments

Please refer to INSTALL.md.

Step2: Prepare datasets

Download the CARGO dataset and modify the dataset path. Line 22, 60, 100 and 140 in cargo.py .

self.data_dir = XXX

Step3: Prepare ViT Pre-trained Models

Download the ViT-base Pre-trained model and modify the path. Line 11 in VDT.yml:

PRETRAIN_PATH: XXX

Training & Testing

Training VDT on the CARGO dataset with one GPU:

CUDA_VISIBLE_DEVICES=0 python3 tools/train_net.py --config-file ./configs/CARGO/VDT.yml MODEL.DEVICE "cuda:0"

Testing VDT on the CARGO dataset:

CUDA_VISIBLE_DEVICES=1 python3 tools/train_net.py --config-file ./configs/CARGO/VDT.yml --eval-only MODEL.WEIGHTS your_model_pth_path MODEL.DEVICE "cuda:0"

Acknowledgement

Codebase from fast-reid. So please refer to that repository for more usage.

[3] Citation

If you find this code useful for your research, please kindly cite the following papers:

@InProceedings{Zhang_2024_CVPR,
    author    = {Zhang, Quan and Wang, Lei and Patel, Vishal M. and Xie, Xiaohua and Lai, Jian-Huang},
    title     = {View-decoupled Transformer for Person Re-identification under Aerial-ground Camera Network},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
    year      = {2024}
}

If you have any question, please feel free to contact me. E-mail: zhangq48@mail2.sysu.edu.cn