This repository contains the code for the article "Data-driven nonsmooth optimization" by S. Banert, A. Ringh, J. Adler, J. Karlsson, and O. Öktem. An arxiv version of the article can be found here.
The code contains the following
- Files used for training the algorithm.
- Files used for validation, both generate one slice/reconstruction and objective function values on a batch of data.
- Files for generalization to deconvolution.
Note that the Mayo Clinic data used in the training do not belong to the authors and must therefore be obtained separately, see here.
Clone this repository, and the ODL repository. Install ODL from source, e.g., by following the ODL installation instructions. Also install numpy, scipy, and ASTRA (version 1.8.3). If you are using conda, the latter can be installed with the following command
- $ conda install -c astra-toolbox astra-toolbox=1.8.3
After this, the scripts can be run using, e.g., spyder. Training has been done using ODL commit 0ab389f, and validation one done using ODL commit eff7129.
Sebastian Banert, Postdoc
Department of Mathematics, KTH Royal Institute of Technology, Stockholm, Sweden
banert@kth.se
Axel Ringh, PhD student
Department of Mathematics, KTH Royal Institute of Technology, Stockholm, Sweden
aringh@kth.se
Jonas Adler, PhD student
Department of Mathematics, KTH Royal Institute of Technology, Stockholm, Sweden
Elekta Instrument AB, Stockholm, Sweden
jonasadl@kth.se
Johan Karlsson, Associate Professor
Department of Mathematics, KTH Royal Institute of Technology, Stockholm, Sweden
johan.karlsson@math.kth.se
Ozan Öktem, Associate Professor
Department of Mathematics, KTH Royal Institute of Technology, Stockholm, Sweden
ozan@kth.se
We acknowledge Swedish Foundation of Strategic Research grants AM13-0049 and ID14-0055, Swedish Research Council grant 2014-5870 and support from Elekta.
The authors thank Dr. Cynthia McCollough, the Mayo Clinic, and the American Association of Physicists in Medicine for providing the data necessary for performing comparison using a human phantom.