French
English
Missions
As part of a project funded by CNRS (call for projects OPEN promoting open source software), the recruited person will be responsible for supporting and developing the open source software RTK for tomographic reconstruction to
- create a wider community of users by offering user documentation associated with Python command-line tools and notebooks,
- improve speed by using alternative software for a number of elementary operations,
- use RTK in reconstruction algorithms using machine learning.
Activities
The required developments are:
- user documentation integrated into the toolkit code.
- implementation of Python command-line tools that could be distributed via Pypi to avoid the compilation step, which is currently necessary and difficult for some users.
- interconnection with other reconstruction software (https://www.astra-toolbox.com/, https://www.ccpi.ac.uk/CIL, https://github.com/CERN/TIGRE, etc) for mutual benefit (speed and new functionalities for RTK),
- possibility of efficiently integrating RTK into a machine learning model, which would require avoiding transfers between CPU and GPU memories. This feature would enable RTK to be interconnected with Pytorch, for example, or with higher-level platforms such as Spyrit, developed at CREATIS.
Skills
Proficiency in Git, CMake, Python and C++. Cuda would be a plus.
Good command of English.
PhD in tomographic reconstruction.
Work Context
RTK is a software toolkit for reconstructing 3D tomographic images from 2D X-ray projections. This calculation is at the heart of medical and industrial scanners. RTK enables fast, parallelized reconstruction on CPU or GPU, of 3D or 4D images (spatio-temporal or spectral), with analytical or iterative algorithms. It can read data from a wide range of medical and industrial scanners. RTK has been developed as a module of ITK, a medical image segmentation and registration software. RTK is a set of C++ and Cuda codes, and can be used to be integrated into a software application, such as the one of a CT scanner. It is also distributed as a Python binary package for use without prior compilation. Command-line tools enable its use without Python or C++ development.