GeoExplorer

PyTorch implementation of GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration (ICCV 2025)

Running

Set-up the environment:

conda env create -f environment.yml

Data preparation:

Please follow the repo of GOMAA-Geo and AirLoc to do the data preprocessing for the Masa, xBD and MM-GAG datasets. For the SwissView dataset, please download the images from the Huggingface Repo SwissView.

1. get patches set path="../data/swissview/swissview100_patches", img_path="../data/swissview/SwissView100/" for SwissView100, and set path="../data/swissview/swissviewmonuments_patches", img_path="../data/swissview/SwissViewMonuments/aerial_view" for SwissViewMonuments.

python get_patches.py

2. get features for areial views login with your Hugging Face token login("HuggingFace_Token_Here"); set data_path="../data/swissview/swissview100_patches/patches/*", save_path="../data/swissview/swissview100_sat_patches.npy"

python get_sat_embeddings.py

3. get features for ground views (SwissViewMonuments only) set data_path="../data/swissview/SwissViewMonuments/ground_view/", save_path="../data/swissview/swissviewmonuments_grd.npy"

python get_grd_embeddings.py

Training and Validation:

Set configurations and parameters in config.py, cfg.dataset == 'masa' for training, cfg.dataset == 'swissview' or cfg.dataset == 'swissviewmonuments' for validation.

To train the model for action-state modeling:

python pretrain.py

To train the model to do curiosity-driven exploration:

python train.py

To run inference, run the following command:

python validate.py

Citation and Acknowledgements

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

@inproceedings{mi2025geoexplorer,
  title={GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration},
  author={Mi, Li and B{\'e}chaz, Manon and Chen, Zeming and Bosselut, Antoine and Tuia, Devis},
  booktitle={ICCV},
  pages={6122--6131},
  year={2025}
}

@inproceedings{sarkar2024gomaa,
  title={Gomaa-geo: Goal modality agnostic active geo-localization},
  author={Sarkar, Anindya and Sastry, Srikumar and Pirinen, Aleksis and Zhang, Chongjie and Jacobs, Nathan and Vorobeychik, Yevgeniy},
  booktitle={NeurIPS},
  pages={104934--104964},
  year={2024}
}