Date: October 17, 2019 (afternoon)
Room: Madrid 2
Joining Mailing List:
Join the AirLab tutorial mailing list to keep up to date. Just send an e-mail to [email protected] with the subject JOIN MAILING LIST
Preparation for Hands-on Session:
If you bring with you your laptop, in the second session you will have the opportunity to try out AirLab with our assistance. As the internet at conferences are usually slow we highly recommend to install PyTorch 1.1 (and CUDA if you have a GPU in your laptop) prior to the tutorial.
A myriad of medical image analysis tasks rely on image registration as a "preprocessing" on which the target model builds on. The development and implementation of such image registration algorithms have been tedious so far and often requires specific programming skills such as parallel computing with CUDA at a low abstraction level. In this tutorial, the novel image registration framework: Autograd Image Registration Laboratory (AirLab) will be introduced. AirLab is a modern framework for image registration which allows rapid-prototyping for image registration algorithms. It is written in Python and utilizes the autograd functionality of PyTorch in order to exempt the developer from providing explicit gradients. The hands-on tutorial will be for beginners and intermediate-level students, experienced researchers and industry delegates. Participants will learn the principles of AirLab, how AirLab can be integrated into a project and how AirLab can be used in their daily research.
GitHub: https://github.com/airlab-unibas/airlab
- Basic concepts and applications
- Novel learning-based concepts
- Main issues with development and implementation
- Current image registration frameworks and tools
- Primer on PyTorch with basic examples
- Basic concept behind AirLab
- Hello world example in AirLab
- Basic features for classic image registration
- How to install AirLab
- Setting up software requirements and environment together
- How to use AirLab
- How to extend AirLab and contribute to the project
- Discussion with participants about the future of AirLab
- AirLab for Learning-based Concepts (on-going)
- Feature requests
Dr. Christoph Jud Department of Biomedical Engineering, University of Basel, Switzerland
Robin Sandkühler, M.Sc. Department of Biomedical Engineering, University of Basel, Switzerland
Robin Sandkühler, Philippe C. Cattin, Christoph Jud
Robin Sandkuehler, Christoph Jud, Simon Andermatt, and Philippe C. Cattin. "AirLab: Autograd Image Registration Laboratory". arXiv preprint arXiv:1806.09907, 2018. link