This is the repo for the CVPR Image Matching Workshop paper A case for using rotation invariant features in state of the art feature matchers. We implement a rotation equivariant LoFTR-version by using steerable CNNs.
Please see the LoFTR repo or the file LoFTR_README.md for instructions on how to obtain the data and run the code.
We add a single dependency, namely e2cnn.
The new config-files configs/loftr/outdoor/loftr_ds_e2_dense*.py contain the parameters used for our SE2-LoFTR experiments.
Models trained on MegaDepth can be found here.
- Implement the rotation equivariant backbone as an
EquivariantModule.
If you find this code useful in your research, please cite our paper as well as the LoFTR and e2cnn papers:
@inproceedings{bokman2022se2loftr,
title={A case for using rotation invariant features in state of the art feature matchers},
author={B\"okman, Georg and Kahl, Fredrik},
booktitle={CVPRW},
year={2022}
}
@article{sun2021loftr,
title={{LoFTR}: Detector-Free Local Feature Matching with Transformers},
author={Sun, Jiaming and Shen, Zehong and Wang, Yuang and Bao, Hujun and Zhou, Xiaowei},
journal={CVPR},
year={2021}
}
@inproceedings{e2cnn,
title={{General E(2)-Equivariant Steerable CNNs}},
author={Weiler, Maurice and Cesa, Gabriele},
booktitle={Conference on Neural Information Processing Systems (NeurIPS)},
year={2019},
}