A PyTorch library for differentiable submodular minimization.
At the moment only one- and two-dimensional graph cut functions have been implemented, so that this package provides differentiable (with respect to the input signal and the weights) total variation solvers.
After installing PyTorch, you can install the package with:
python setup.py install
To run the tests you simply have to run:
python setup.py test
- [DK17] J. Djolonga and A. Krause. Differentiable learning of submodular models. In Advances in Neural Information Processing Systems (NIPS), 2017.
- [NB17] V. Niculae and M. Blondel. A regularized framework for sparse and structured neural attention. arXiv preprint arXiv:1705.07704, 2017.