carterbox/torch-radon is a fork of matteo-ronchetti/torch-radon with some
modules removed (shearlets, reconstruction) and the build system replaced with
the extension system from PyTorch. This fork is maintained separately because
the upstream project is unmaintained. If the upstream project becomes active
again, this fork will attempt to merge its improvements upstream.
TorchRadon is a PyTorch extension written in CUDA that implements differentiable routines for solving computed tomography (CT) reconstruction problems.
The library is designed to help researchers working on CT problems to combine deep learning and model-based approaches.
Main features:
- Forward projections, back projections and shearlet transforms are
differentiable and integrated with PyTorch
.backward(). - Up to 125x faster than Astra Toolbox.
- Batch operations: fully exploit the power of modern GPUs by processing multiple images in parallel.
- Transparent API: all operations are seamlessly integrated with PyTorch,
gradients can be computed using
.backward(), half precision can be used with Nvidia AMP. - Half precision: storing data in half precision allows to get sensible speedups when doing Radon forward and backward projections with a very small accuracy loss.
Implemented operations:
- Parallel Beam projections
- Fan Beam projections
- 3D Conebeam projection
TorchRadon is much faster than competing libraries:
See the Tomography Benchmarks repository for more detailed benchmarks.
Currently only Linux is supported. Windows not supported mainly because there not yet a Windows package for PyTorch on the conda-forge channel.
Please read about how to setup and use the conda package manager before attempting the following command.
conda install --channel conda-forge carterbox-torch-radonNo PYPI packages will be provided because pip was not designed for mixed-language software distribution.
If you are using TorchRadon in your research, please cite the following paper:
@article{torch_radon,
Author = {Matteo Ronchetti},
Title = {TorchRadon: Fast Differentiable Routines for Computed Tomography},
Year = {2020},
Eprint = {arXiv:2009.14788},
journal={arXiv preprint arXiv:2009.14788},
}
Install testing dependencies with pip install .[testing]
then test with:
pytest tests/