1.5.2

Considered feature complete. I'll support existing implemented features and try to fix them if they break (or are already broken).

New features will be at the responsibility of the implementer.

I might update once in a while, but mostly to practise and get up to date on new C++ features, e.g. C++20 has math constants, concepts and a few other cool things.

The binary is for Windows 10 with CUDA (requires Nvidia GPU) and tuned for AVX2 which is in most modern processors.

1.5.1 'Stable' release with OpenCL and CUDA option

Changelog:

• Scatter correction re-worked, no more "correction factor"
• OpenCL forward projection option
• DICOM RS structure support for cropping and WEPL re-calculation
• Update from ITK 4 to 5 (Plastimatch and RTK at master branch)
• Major revisions to code safety, with clang-tidy, PVS Studio, ReSharper, CppCheck
• Major revisions to code format, using LLVM style with clang-format
• Major work on CI, using gitlab-runner to test the code (more 'unit' tests still needed)
• Separation of compute and user interface (still needs some work)
• Plastimatch will now use OpenMP instead of CUDA for registrations (major performance gain on some systems)
• OpenCL version almost achieves the same performance as the CUDA version.

Important: Must be installed on a location where the user has write permissions (i.e. NOT "Program Files").

System requirements:

• The software itself requires approximately 804 MB of free disk space.
• The ram usage depends mostly on the number of projections (Up to three equivalents of the downscaled projections are held in RAM at the same time), but more than 8 GB is recommended.
• CPU must be 64-bit and for these installers AVX2 capable.
• The screen resolution should be larger than 1161 by 747 pixel.
• Arrow keys + PageUp and PageDown are required for manual registration.

Despite having a major version of 1, this is still very much a work in progress and API breaking changes will happen within major versions.

The binaries are optimized with the /arch:AVX2 flags. Please, let me know if you need a non-AVX2 optimized installer. I am only able to compile 64-bit versions.

Please report any issues/bugs, either as issues here on GitHub, or by email: agravgaard@protonmail.com

First working installer (64-bit, CUDA)

This is the first release with a working installer

This version requires a CUDA capable graphics card with a driver supporting code generated by CUDA 9.1 (R390 or newer).

The code is optimized with the /arch:AVX2 /favor:INTEL64 flags. Please, let me know if you need a non-AVX2 or /favor:AMD64 optimized installer. I am only able to compile 64-bit versions.

Important: Must be installed on a location where the user has write permissions (i.e. NOT "Program Files").

The software requires approximately 1 GB of free disk space.
The ram usage depends mostly on the number of projections (Up to three equivalents of the downscaled projections are held in RAM at the same time), but more than 8 GB is recommended.
CPU must be 64-bit and for this installer AVX2 capable.
The GPU must be compatible with a nVidia driver of version R390 or newer.
The screen resolution should be larger than 1161 by 747 pixel.
Arrow keys + PageUp and PageDown are required for manual registration.

Despite having a major version of 1, this is still very much a work in progress and API breaking may happen within major versions. However, this is not intended as an API anyway.

Please report any issues/bugs, either as issues here on GitHub, or by email: agravgaard@protonmail.com