PyTorch implementation of Radon transform. Right now only 2-dimentional case on CPU is supported. Contributions to higher dimentional cases and GPU cases are welcome.
The motivation of this project is the disagreement of the inverse radon transform in scikit-image implementation with MATLAB (refer to issue #3742).
The requirements for the conda environment in which I have tested this code are started in requirements.txt. The main dependencies are
1. python >= 3.6.2
2. torch >= 1.0
3. numpy >= 1.16.2
4. scipy >= 1.2.1
5. scikit-image >= 0.16.2
python setup.py install
some examples are provided in tests. The results from phantom image are presented in following figures.
The comparison with the signogram and reconstruction from MATLAB on phantom image is provided by test_comparison.py:
sinogram rms error: 0.42, mae error: 0.0169
FBP rms error: 0.00805, mae error: 0.00601
MIT License see LICENSE.