UAV-geoloc

Code for Deep learning based cross-view image matching for UAV geo-localization.

Getting Started

Installation

1. Clone the repo

git clone https://github.com/LuoXubo/UAV-geoloc

2. Install requirements

conda env create -f environment.yml
conda activate uavgeoloc

Data Preparation

1. Download the UAVDT dataset and extract it to data/UAV-benchmark-M/.
2. Arrange the dataset as follows:

UAV-geoloc
├── datasets
│   ├── query
│   │   ├── 1
│   │   │   ├── 000001.jpg
│   │   │   ├── ...
│   ├── database
│   │   ├── 1
│   │   │   ├── 000001.jpg
│   │   │   ├── ...

Testing

Test the model with the following commands:

1. Coarse localization

cd Coarse

python3 test.py \
--name='final_three_view_long_share_d0.75_256_s1_google_LPN4_lr0.001' \
--batchsize=128 \
--gpu_ids='0'

python demo_all.py

cd Refine

python dfm_selected_top1.py

Citation

If you find this work useful, please consider citing the following paper:

@INPROCEEDINGS{10137193,
  author={Luo, Xubo and Tian, Yaolin and Wan, Xue and Xu, Jingzhong and Ke, Tao},
  booktitle={2022 International Conference on Service Robotics (ICoSR)},
  title={Deep learning based cross-view image matching for UAV geo-localization},
  year={2022},
  volume={},
  number={},
  pages={102-106},
  keywords={Location awareness;Deep learning;Satellites;Service robots;Image matching;Refining;Lighting;deep learning;image matching;geo-localization;autonomous drone navigation},
  doi={10.1109/ICoSR57188.2022.00028}}

Acknowledgements

This repo is built upon the LPN and DFM. We thank the authors for sharing their code.