This tomography code is based on the FASTA implementation of the Fast Forward-Backward Splitting algorithm and the ASTRA library for forward and backward ray-transform operators. The FASTA paper talks about all the implementation details like adaptive stepsizes, accelerated convergence, and so on.
- install ASTRA from source to enable GPU capabilities
- install pytorch (tested with pytorch v1.5 and 1.6)
- install the Operator Discretization Library ODL conda install -c odlgroup odl
- 'python setup.py install' in this folder
- [optional] 'python setup.py develop' for developer mode
- run ./tests/test_kates_data.py or ./tests/test_kates_data_multiscale.py
- interface Total General Variation regularization from the CCPi regularisation toolkit and write a Plug-and-Play regularizer to use in FASTA
- port subpixel registration to the GPU, this is probably easiest with the CuPy package, since it has exactly the same interface as numpy. The code is in ./fastatomography/util/register_translation.py, and is copied from skimage
- implement higher-order TV regularisation, see above