dpisanti/mbl-mars

Pretrained Weights for Map-based Localization (MbL) on Mars

Pretrained model weights for the observation-to-map matchers used in the MbL pipeline proposed in:

Geometry-aided Vision-based Localization of Future Mars Helicopters in Challenging Illumination Conditions
Dario Pisanti, Robert Hewitt, Roland Brockers, Georgios Georgakis
[arXiv] · [Code]

Weights

The weights_loftr.zip archive corresponds to the weights/loftr/ subfolder expected by the mbl_mars pipeline. Extract it and place the resulting loftr/ folder under weights/ in the repository root.

Path (after extraction)	Input	Description
weights/loftr/V1/geo/	Image triplet (observation, map patch, depth)	Geo-LoFTR model trained on MARTIAN data with HiRISE-like maps (0.25 m/px resolution)
weights/loftr/V1/finetuned/	Image pair (observation, map patch)	Original LoFTR model fine-tuned on MARTIAN data with HiRISE-like maps (0.25 m/px resolution)
weights/loftr/pretrained/	Image pair (observation, map patch)	Off-the-shelf LoFTR model trained on MegaDepth (see LoFTR repo)
weights/loftr/geo_ctx/ (experimental)	Image triplet (observation, map patch, depth)	Geo-LoFTR fine-tuned on CTX-like maps (6 m/px resolution) for lower-resolution map matching

Training data was generated using MARTIAN, a simulation framework that uses real HiRISE orbital imagery of the Jezero crater.

Usage

Download weights_loftr.zip, extract it, and place the loftr/ folder under weights/ in the root of the mbl_mars repository:

git clone https://github.com/nasa-jpl/mbl_mars.git

Then follow the setup and usage instructions in the mbl_mars README.

Citation

If you find this work useful, please cite:

@misc{pisanti2026geometryaidedvisionbasedlocalizationfuture,
      title={Geometry-aided Vision-based Localization of Future Mars Helicopters in Challenging Illumination Conditions}, 
      author={Dario Pisanti and Robert Hewitt and Roland Brockers and Georgios Georgakis},
      year={2026},
      eprint={2502.09795},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2502.09795}, 
}

Copyright

Copyright 2026, by the California Institute of Technology. ALL RIGHTS RESERVED.
United States Government Sponsorship acknowledged.
Any commercial use must be negotiated with the Office of Technology Transfer at the California Institute of Technology.

This software may be subject to U.S. export control laws.
By accepting this software, the user agrees to comply with all applicable U.S. export laws and regulations.
User has the responsibility to obtain export licenses, or other export authority as may be required
before exporting such information to foreign countries or providing access to foreign persons.