CAMERA Workshop: Algorithms for Tomographic Reconstruction: State-of-the-Art and Future Goals:
Nov. 9-11, 2016, Lawrence Berkeley National Laboratory
Organizers: D. Parkinson (ALS-CAMERA), D. Gursoy (APS-Argonne), K. Champley (LLNL)
This workshop will focus on the current state-of-the-art for tomographic reconstruction algorithms. The goal is to bring together users, practitioners, and developers to asses the current landscape of available algorithms, to investigate communalities and differences among the various techniques, and to discuss a range of topics, from required theoretical and algorithmic advancements on through to practical issues of implementation and deployment.
We are bringing together developers and users of synchrotron tomography software. The workshop will have two kinds of talks:
(1) Technical talks from developers to describe their current algorithms and capabilities, and their future potential.
(2) Talks from users of tomography software to discuss successes and unmet needs, and to find common goals.
We plan to include ample time for discussion and hacking, with our software people available to help get software packages running on various computer systems at LBNL if developers so choose (including our high performance computer NERSC, our GPU clusters, and standard beefy workstations). We are also happy to provide developers with access to a wide variety of challenging data sets--we have been archiving all raw data from the ALS hard X-ray tomography beamline for the past 5 years.
We plan to hold the following sessions during the workshop:
Algorithms for Synchrotron Tomography
Synchrotron tomography experiments continue to push the limits of time and spatial resolution, while at the same time constantly expanding the kinds of in situ environments that are used for samples. In addition, because of the tight schedule between users and the limited beam time for each user, it is often difficult to spend the optimal amount of time aligning, cleaning, and adjusting the hardware. Each of these factors can contribute to challenges in obtaining a tomographic reconstruction that lets a user see what they need to see--whether by eye or through a subsequent processing or analysis algorithm. From another perspective, the rapidly increasing data rates combine with in situ experiments to frequently require very fast feedback, which requires maximizing the speed of reconstruction. This session is devoted to the latest advances in algorithms that improve the speed and/or the image quality of tomographic reconstruction, whether for a particular situation or for an array of applications.
Roundtable discussion on benchmarking
A large number of both algorithms and software packages for tomographic reconstruction are now available. In the scientific literature (as well as in ads for commercial software), claims of novelty, quality, and speed, are frequently made. In some cases comparisons with a small sample of other methods are published, while in other cases various speed or quality claims are simply stated. This roundtable discussion will invite a few leading contributors to synchrotron tomography software and algorithms to discuss their views on how the community might agree on some appropriate ways to fairly evaluate and test new algorithms and software. Audience members will have a chance to submit questions in advance, and there will also be a chance for questions made directly to the panel near the end of the session.
Software for Synchrotron Tomography
A key feature of synchrotrons is that they are generally user facilities. Due to the enormous expansion of the applicability of synchrotron tomography in a number of fields, many of the users of tomography beamlines are not experts in synchrotron science or tomography, but in biology or materials science or another field. Because of this, many beamline scientists and synchrotron software teams have been striving to provide users not just algorithms, but with complete software packages that package these algorithms in a way that makes them useful to this broad array of users. This session is devoted to the developers of software for synchrotron tomography, giving them a chance to present the key motivators and features of their software, and their future plans. These presentations may lead to ideas for new collaborations or to ways of sharing and integrating features between packages so that they benefit each other.
Hands-on with Algorithms and Software
A key goal of this workshop is to get an overview of the various algorithms and software packages that are available for processing tomographic data, to allow users and beamline scientists to get some hands-on experience with these packages, and to allow algorithm and software developers to become more familiar with other available algorithms and software. We believe this will lead to less duplication of effort, which in turn will allow more effort to be spent on developing truly new and unique capabilities. Each developer will be assigned a time slot during which they will host small groups of participants who want some “hands on” time with the software being developed. We will work with all interested developers in advance to deploy the software on an appropriate computer system at LBNL, and make accounts available on these systems for all participants who want to try the software. Available systems include:
Cori Phase I at NERSC ( http://www.nersc.gov/systems/cori/ ), a supercomputer with 52,160 compute cores.
Linux, Windows, and Mac Workstations (with GPU)
We will also collect challenging (and not-so-challenging) data sets from multiple facilities in advance and share them with developers so that the participants can more easily evaluate and understand the applicability of particular algorithms and software packages.
Algorithms and Software from a User's Perspective: Success stories and unmet needs
During this session, users of tomography software and algorithms (including both beamline scientists and facility users) will discuss their experiences. They will share success stories, including overcoming challenges in extracting information, or in increasing speed to provide needed feedback during experiments. They will also share their as yet unmet needs, or places in which algorithms or software are not performing as well as they would like. Finally, speakers in this session will take a few minutes each to highlight what they found most interesting, exciting, or promising from the software and algorithm developer presentations, especially if they had a chance to try something new during the hands-on session.
Dula Parkinson, CAMERA/LBNL
James Sethian, CAMERA/LBNL
Daniel Pelt, CAMERA/LBNL
Singanallur Venkatakrishnan, CAMERA/LBNL
Doga Gursoy, APS/Argonne
Francesco De Carlo, APS/Argonne
Rafael Vescovi, APS/Argonne
Francesco Brun, ESRF
Andreas Kopmann, KIT
Nghia Vo, Diamond
Kyle Champley, LLNL
Yijin Liu, SSRL/SLAC
Peter Ercius, NCEM/LBNL
Ryan Crum, LLNL
Willem Jan Palenstijn, CWI
Tekin Bicer, Argonne
Darren Thompson, CSIRO
Michael Sutherland, DMEA
Axel Ekman, UCSF/LBNL
Nicola Wadeson, Diamond
Alex Hexemer, CAMERA/LBNL
Holger Kohr, CWI
Mark Basham, Diamond
Xiao Wang, Purdue
Nikolay Malitsky, BNL
Simo Makiharju, UC Berkeley
Felix Beckmann, HZG, Germany
Viktor Nikitin, Max iV
Schedule of Talks:
Please send mail to