CERN / TIGRE

TIGRE: Tomographic Iterative GPU-based Reconstruction Toolbox
BSD 3-Clause "New" or "Revised" License
573 stars 188 forks source link

gpuarray supported forward and backprojectors #300

Open AnderBiguri opened 3 years ago

AnderBiguri commented 3 years ago

I have code from a contributor of TIGRE that allows, for users with the distributed computing toolbox, and for CT images that fit inside the GPU together with the algorithm, to avoid CPU-GPU transfers and to everything in the GPU.

I don't have the physical time to add them to TIGRE, but if someone wants to, happy to share it and give a hand on the design on how to integrate it seamlessly into current TIGRE.

yliu88au commented 3 years ago

Good move! For python version, I would suggest Cupy (Numpy equivalent) package.

tsadakane commented 3 years ago

@AnderBiguri Is this related to Collision avoidance trajectories for on-line trajectory optimization in C-arm CBCT by S. Hatamikia et al. of Fully3D 2021? Just out of curiosity.

AnderBiguri commented 3 years ago

Hi @tsadakane , it is not. In that work, a full C++ implementation was coded, particularly optimized for the size and amount of projections of the application.

The code I have is from Matt Jacobson, I believe it will be presented at:

Ordered Subsets PWLS Reconstruction of Multi-Layer Detector MV-CBCT Data Using TIGRE M Jacobson1*, M Lehmann2, P Huber2, M Myronakis1, T Harris1, I Valencia Lozano1, R Fueglistaller2, C Williams1, D Morf2, R Berbeco1, (1) Brigham and Women's Hospital, Dana-Farber Cancer Institute, and Harvard Medical School, Boston, MA (2) Varian Medical Systems, Baden-Dattwil, Switzerland AAPM 2021