In this talk we present ODL (Operator Discretization Library, https://github.com/odlgroup/odl), a Python framework for fast prototyping in inverse problems. It features classes and functions that closely resemble mathematical structure, for instance an "Operator" class with "derivative" and "adjoint", or "Functional" with "gradient", "convex_conj" and "proximal". This allows to represent and solve inverse problems on a high level, while harnessing the computational power of optimized libraries for subtasks. We will demonstrate the concept on practically relevant cases.

This presentation is part of Minisymposium “MS76 - Solving Inverse Problems in minutes: Software for imaging (2 parts)”
organized by: Ozan Öktem (KTH - Royal Institute of Technology) , Holger Kohr (Thermo Fisher Scientific) , Jonas Adler (KTH Royal Institute of Technology) .

Authors:

Holger Kohr (Thermo Fisher Scientific)

Jonas Adler (KTH Royal Institute of Technology)

Ozan Öktem (KTH - Royal Institute of Technology)

Keywords:

computed tomography, deep learning, image reconstruction, inverse problems, nonlinear optimization, numerical linear algebra, numerical software