Awesome

HVSA

Official PyTorch implementation for Hypersphere-Based Remote Sensing Cross-Modal Text–Image Retrieval via Curriculum Learning.

News :tada:

• 📣 Oct 2023 - The code of HVSA has been released.
• 📣 Oct 2023 - Codes is coming soon (before Nov).
• 📣 Sep 2023 - Paper Accepted by TGRS.

Dependencies

Download the dataset files and pre-trained models. Files about seq2vec are from here.

Install dependencies using the following command.

pip install -r requirements.txt

Dataset preparation

All experiments are based on RSITMD and RSICD datasets,

RSICD

We followed the same split provided by AMFMN. Dataset splits can be found in data/rsicd_raw.

RSITMD

We followed the same split provided by AMFMN. Dataset splits can be found in data/rsitmd_raw.

Train

# RSITMD Dataset
python train.py --config configs/HVSA_rsitmd.yaml

Evaluate

# RSITMD Dataset
python eval.py --config configs/HVSA_rsitmd.yaml

Performance

Citing HVSA

If you find this repository useful, please consider giving a star :star: and citation:

@ARTICLE{10261223,
  author={Zhang, Weihang and Li, Jihao and Li, Shuoke and Chen, Jialiang and Zhang, Wenkai and Gao, Xin and Sun, Xian},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Hypersphere-Based Remote Sensing Cross-Modal Text–Image Retrieval via Curriculum Learning}, 
  year={2023},
  volume={61},
  number={},
  pages={1-15},
  doi={10.1109/TGRS.2023.3318227}}

Acknowledgment

The implementation of HVSA relies on resources from <a href="https://github.com/fartashf/vsepp">VSE++</a>, and <a href="https://github.com/xiaoyuan1996/AMFMN">AMFMN</a>. We thank the original authors for their open-sourcing.