ToMoBAR is a Python library of direct and model-based regularised iterative reconstruction algorithms with a plug-and-play capability for parallel-beam geometry. ToMoBAR offers you a selection of various data models and regularisers resulting in complex objectives for tomographic reconstruction. ToMoBAR can operate on a GPU device in a device-to-device fashion (in-memory) using CuPy API, therefore ensuring a better computational efficiency. This is especially relevant for iterative algorithms where multiple projection/backprojection and image filtering (e.g. regularisation) operations are in place.

With the GPU device controlling API exposed, it can also support multi-GPU parallel computing. This functionality is exploited by the HTTomo UI software through MPI protocols and parallel HDF5.

Installation

Please see the detailed instructions here.

Usage

Please see various tutorials on direct and iterative reconstruction.

Benchmarks

As ToMoBAR relies on ASTRA toolbox modules, one can expect a similar to ASTRA performance. The speed-up (up to 5 times) can be achieved by using CuPy implementations of the advanced iterative methods with regularisation (see FISTA), instead of the host-device-host versions.