CCP SyneRBI and CCPi are co-organising a workshop + hackathon at UCL (London, UK) to enable integration of various software libraries specialising in physics-modelling for image reconstruction with Python deep learning frameworks such as PyTorch. 

Attendance information is below. 

Contact tomography@stfc.ac.uk if you are interested in taking part to the event.

Workshop: Integration of libraries for physics informed deep learning in imaging 

The workshop will be hybrid: 

• UCL Foster Court UCL Estates: Building Location: FOSTER COURT, Room 215.
  Contact tomography@stfc.ac.uk if you are interested in participating in person. 

• Remote access:
  Teams meeting link, Meeting ID: 331 164 268 444, Passcode: QD7zB629  

Schedule

Monday 7 April: 13:30-18:00 BST 

All talks are 15 min + 5 min discussion 

• Overview of STIR/SIRF architecture and need for this workshop and following hackathon (Kris Thielemans, UCL, London, UK)
• SIRF pytorch interface on Python level (Imraj Singh, UCL, London, UK)
• Parallelproj (Georg Schramm, KU Leuven, Belgium)
• Overview of CIL architecture (Gemma Fardell, STFC, London, UK)
• Overview of interfacing C++/CUDA with Python (Casper da Costa-Luis, STFC, UK)
• CuVec and NiftyPET (Pawel Markiewicz & Casper da Costa-Luis, STFC, UK)

Break (20 min) 

• Interfacing GPU image data with other Python packages in RTK (Simon Rit, CREATIS, Lyon, France)
• Integrating ASTRA and Python software (Willem-Jan Palenstijn, LIACS, Leiden, Netherlands)
• Challenges and solutions when combining ODL, PyTorch and Astra (Justus Sagemüller, KTH, Stockholm, Sweden)

Discussion (30 min) 

Hackathon: Integration of STIR/SIRF/CIL with pytorch 

The hackathon will be mostly in-person, accommodating remote attendance as much as possible. Contact tomography@stfc.ac.uk if you are interested in participating, specifying in-person or remote attendance and join our Discord server  

Location:

• UCL Foster Court UCL Estates: Building Location: FOSTER COURT, Room 215. 
• Remote meetings details will be put on the Discord channel. 

Aim: optimise usage of CUDA in STIR/SIRF/CIL when interfacing to PyTorch 

Example tasks: 

• STIR 

• • Allow use CUDA managed pointers for image objects 
  • Add CUDA numerical operations ? 
  • Adjust STIR’s parallelproj interface to use above 
  • Expose via SWIG to python 

• SIRF 

• • Pass through STIR image objects appropriately to python 

• CIL 

• • Investigate DataContainer agnostic on backend, e.g. numpy, cupy, pytorch
  • Efficient data exchange with GPU backend, TIGRE/ASTRA 
  • making a denoising application with torch function 
  • run CIL Algorithm with DL denoising step 

  

Schedule

Tuesday: 9:00-18:00
	introduction and separation into groups 9:00 – 9:45  Group hacking 9:45 – 11:30  Group activity reporting: 11:30 – 12:00  Lunch 12:00 – 13:00  Group hacking 13:00 – 17:00  Group activity reporting: 17:00 – 18:00
Wednesday: 9:00-13:30
	Group hacking 9:00 – 11:30  Final reporting and outlook: 11:30 – 12:30  Optional Lunch 12:30 – 13:30