An adaptive optimization package for the 1d-TV problem for research purpose, proposed in the following paper:
Hamza Cherkaoui, Jeremias Sulam and Thomas Moreau "Learning to solve TV regularised problems with unrolled algorithms", accepted at NeurIPS 2020.
This package implements:
- Classical solver for TV-regularized optimization problems: (primal-dual) ISTA, (primal-dual) FISTA, Condat-Vu.
- Learnable algorithms: All the iterative versions cited above.
- Official source code repo: https://github.com/hcherkaoui/carpet
The required dependencies to use the software are:
- Numpy >= 1.14.0
- Scipy >= 1.0.0
- Joblib >= 0.11
- Torch >= 1.4.0
- Matplotlib >= 2.1.2
- Prox_tv
All material is Free Software: BSD license (3 clause).
In order to perform the installation, run the following command from the carpet directory:
python3 setup.py install --user
To run all the tests, run the following command from the carpet directory:
pytest
You can check the latest sources with the command:
git clone git:https://github.com/hcherkaoui/carpet
or if you have write privileges:
git clone [email protected]:hcherkaoui/carpet