TIGRE 2.2

Features

• You can now select which GPUs to run the code on, either by name, or GPUid.
• Improved compilation, now there is no need to edit setup.py/xml files for your GPU/CUDA SDK
• Unify source: there is only one folder with all the CUDA code for both python and MATLAB.
• Detector rotation now supported on parallel beam
• Add NikonDataLoader() to python.
• Code quality optional code added, to be able to maintain decent python.
• Demos for python available
• Add 3D Shepp-Logan head phantom to Python.
• Replace noise generator with CUDA implementation.

BugFix

• Fix bug where compute 5.0 was not being compiled in python
• Some arbitrary rotation angles were not being passed to the kernel
• Reduced memory needed in MATLAB NikonDataLoader
• Mayor bug that impedes many iterative algorithms to work in python. The way order_subsets returns arrays changed, now they are not dtype=object, but dtype=float
• ComputeV errored for triplet angle inputs, fixed
• Fixed many python algorithm bugs (lost track of some)
• Removed many unnecesary files
• Fix and refactor checkDevices
• Fix backprojection type variable name and its default value in Python.