Awesome

ConGeo

Official method implementation for the ECCV24 paper: ConGeo: Robust Cross-view Geo-localization across Ground View Variations. arxiv

Introduction

ConGeo is a learning framework that can be applied to base Cross-View Geo-Localization (CVGL) architectures for robust CVGL across ground view variations using a single model.

Installation and Get Started

Required environments:

• Linux
• Python 3.7+
• PyTorch 1.10.0+
• CUDA 9.2+
• GCC 5+

Install: Please follow the following steps for installation.

git clone https://github.com/eceo-epfl/ConGeo.git
cd ConGeo
pip install -r requirements.txt

Usage:

We take the CVUSA for example to illustrate the usage of this repo:

• Set the path in dataset and training configurations to the dataset path.
• Train by running:

python train_congeo_cvusa.py

• Eval by running:

python eval_cvusa.py

Tips:

• Change the "train_fov" configuration in the training code to customize your training mode:

0.0: north-aligned, value between (0.0, 360.0): limited FoV, 360.0: arbitrary orientations

• Change the "fov" configuration in the eval code to change evaluation settings:

0.0: north-aligned, value from (70.0, 90.0, 180.0): limited FoV, 360.0: arbitrary orientations

• Set "train_fov=360", and "random_fov=True" can enable training with random FoVs between (70, 360) degrees

Citation and Acknowledgement:

We would like to thank the authors of Sample4Geo for the code basis of this work. If you find this work helpful, please consider citing:

@article{mi2024congeo,
  title={ConGeo: Robust Cross-view Geo-localization across Ground View Variations},
  author={Mi, Li and Xu, Chang and Castillo-Navarro, Javiera and Montariol, Syrielle and Yang, Wen and Bosselut, Antoine and Tuia, Devis},
  journal={arXiv preprint arXiv:2403.13965},
  year={2024}
}

@inproceedings{deuser2023sample4geo,
  title={Sample4geo: Hard negative sampling for cross-view geo-localisation},
  author={Deuser, Fabian and Habel, Konrad and Oswald, Norbert},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={16847--16856},
  year={2023}
}