dkazanc/ToMoBAR — Gitpedia

<table> <tr> <td> <div align="left"> <img src="docs/source/_static//tomobar_logo.png" width="450"><br> </div> </td> <td> <font size="5"><b> TOmographic MOdel-BAsed Reconstruction software <a href="https://github.com/dkazanc/ToMoBAR/tree/master/docs/Kazantsev_CT_20.pdf">PAPER (CT Meeting 2020)</a></b></font> <br><font size="3" face="verdana" color="green"><b> ToMoBAR</b> is a Python library of fast direct and model-based regularised iterative algorithms with a plug-and-play capability for reconstruction of parallel-beam geometry data. ToMoBAR offers you a selection of various data models and regularisers resulting in complex objectives for tomographic reconstruction. As ToMoBAR relies on device-to-device methods operating on CuPy arrays it offers significant speed-ups. It also can handle multi-GPU parallel reconstruction through the <a href="https://diamondlightsource.github.io/httomo/">HTTomo</a> framework for big-data processing and reconstruction. ToMoBAR is used in production at <a href="https://www.diamond.ac.uk/Home.html">Diamond Light Source</a>.</font></br> </td> </tr> </table>

Master	Anaconda binaries

Anouncements:

$\sf\color{red}!$ Starting from version 2026.3.0.0, iterative reconstruction methods will no longer be accessible through the RecToolsIR interface and RecToolsIRCuPy should be used instead. The dependency on the Regularisation Toolkit is dropped in favour of the internal CuPy routines. Please see more information in CHANGELOG

CHANGELOG:

See CHANGELOG for all detailed changes.

ToMoBAR highlights:

Check what ToMoBAR can do. Please also see Tutorials and Demos.
ToMoBAR

Installation

Please check the detailed installation guide where all software dependencies are listed.

Software includes:

Wrappers around ASTRA-toolbox to simplify access to various reconstruction methods available in ASTRA-Toolbox
CuPy driven forward/backward projectors to enable faster device-to-device operations and all-in-GPU memory prototyping of algorithms.
Optimised CUDA/CuPy implementation of the fast Log-Polar (Fourier-based) direct reconstruction method.
Regularisation modules that can be used for denoising or for regularisation in iterative methods.
Regularised iterative ordered-subsets FISTA reconstruction algorithm with linear and non-linear data fidelities.
Regularised iterative ordered-subsets ADMM reconstruction algorithm. Very fast, especially with warm start, relaxation and ordered-subsets enabled.
Demos to reconstruct synthetic and also real data