HIERARCHICAL ATTENTION FUSION FOR GEO-LOCALIZATION

February 25, 2021 ยท View on GitHub

Introduction

HAF extracts the multi-scale feature maps from a convolutional neural network (CNN) to perform hierarchical attention fusion for image representations. This repo is the PyTorch implementation of ICASSP2021 paper "HIERARCHICAL ATTENTION FUSION FOR GEO-LOCALIZATION" [pdf]

Installation

Please find detailed steps Here for installation and dataset preparation.

Train & Test

Please find details Here for step-by-step instructions.

Model Zoo

Please refer to Here for trained models.

Inference on a single image

Please refer to Here for inference on a single image.

Train on customized dataset

Please refer to Here to prepare your own dataset.

License

HAF is released under the MIT license.

Citation

If you find this repo useful for your research, please consider citing the paper

@inproceedings{yan2021densernet,
    title={Hierarchical Attention Fusion for Geo-localization},
    author={Liqi Yan, Yiming Cui, Yingjie Chen, Dongfang Liu},
    booktitle={ICASSP}
    year={2021},
}

Acknowledgements

We truely thanksful of the following piror efforts in terms of knowledge contributions and open-source repos. Particularly, "ASLFeat" has a similar approach to ours but using strong supervision.