MBIRTorch: High-performance tomographic reconstruction#
MBIRTorch is a Python package for Model Based Iterative Reconstruction (MBIR) of images from tomographic data.
Key features:
Fast, robust convergence using Vectorized Coordinate Descent [A1].
Supports a wide and increasing range of geometries (see Geometry Models).
Supports integration of AI models using the Plug-and-Play priors framework [A6] [A5].
Automatic parameter selection, with fine-tuning using intuitive meta-parameters.
4D reconstruction of a moving object from a single continuous scan [A4].
Modular, extensible, easy-to-use, object-oriented Python interface.
Fast, portable, seamless use on CPUs or GPUs through the use of PyTorch.
Reconstructions can be performed in just a few lines of python code.
Vectorized Coordinate Descent produces high-quality images quickly.
Based on PyTorch, MBIRTorch can easily run on CPU or GPU.