Self-supervised End-to-End Landmark Detection and Matching

Code for the paper An end-to-end deep learning approach for landmark detection and matching in medical images. The full text of the paper is available at https://arxiv.org/pdf/2001.07434v1.pdf.

Note, the implementation for generating ground truth labels (function get_labels in loss.py) required for self-supervised training of landmark matches is similar to (and inspired by) the method mentioned in End-to-end learning of keypoint detector and descriptor for pose invariant 3D matching.

Usage

Due to restrictions in sharing the medical data used in the paper, the repository is modified to train on CelebA dataset. To train on CelebA dataset, download CelebA dataset from the following link: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html. Clone the repository and install dependencies using requirements.txt. In the root folder of the repository, run the following command on terminal:

python train.py -data_dir <path to CelebA dataset>

If you find the code useful, please cite the following paper:

@inproceedings{grewal2020end,
  title={An end-to-end deep learning approach for landmark detection and matching in medical images},
  author={Grewal, Monika and Deist, Timo M and Wiersma, Jan and Bosman, Peter AN and Alderliesten, Tanja},
  booktitle={Medical Imaging 2020: Image Processing},
  volume={11313},
  pages={1131328},
  year={2020},
  organization={International Society for Optics and Photonics}
}