FLAGSHIP - Initial pre-kickoff meeting and CCP core integration process: plus follow-on meetings
Introductions and roles of related participants.
Conventional X-ray imaging entails a polychromatic X-ray source (i.e. with a full spectrum of energies) but with energy-indiscriminate detectors (registering a single grey-scale channel). However, technological breakthroughs in energy-sensitive detectors enable a new era of tomographic imaging in 'colours' (multiple channels).
Official EPSRC grant notification
http://gow.epsrc.ac.uk/NGBOViewGrant.aspx?GrantRef=EP/P02226X/1
The CCPi SLA and the Flagship grant developers, will both have governance through the Working Group - with these meetings planned every six months (approx June and December).
- Techy-Admin meeting started with this early meeting on 20 March 2017 at RAL RCaH (att 14 +3 remote via webex)
- Start date is 1 April 2017 and end date is 31 March 2020
- Next Working Group meeting is: 12 June 2017 https://www.eventbrite.co.uk/e/ccpi-working-group-meeting-tickets-32227704933
- RSE Conference: http://rse.ac.uk/rse-conference/ 7-8 September 2017 Call for papers deadline is 28 April 2017
- Next Techy-Admin meeting (t-conf: 11 May 2017, informal at Working Group/Fringe 12/13 June 2017, next 12 September 2017 then ...)
The aim for these exchange Techy-Admin meetings are to focus on the "user needs/outcomes" as well as the "techy/maths problems".
Original proposal diopocument:
EP-P02226X-1proposalPartA.pdf and Gantt chart:
EP-P02226X-1proposalPartB.pdf
From the Project Gantt chart there will be a three phase structure (colours on chart) to focus on published user impact.
- BLUE Algorithmic development of a specific user proble (DK, JJ)
- RED CCPi harding proces and integration (SN, EP, EY)
- GREEN User publication of outcomes with links to facilities and lab as appropriate (MB, WK, BC)
This will be repeated creating a set of beta versions etc., of the toolkit.
Presentations CCPi Flagship Grant: Phil Withers
Invited External Speakers
Presentations Core CCPi development intro and updates
CCPi Flagship Management Process
CCP admin web links etc.
- Create a tab on the main CCPi webpage
- Use project planning tools - e.g. slack (private messaging), trello (public project outcomes) and github/ccpforge site.
- Publicity and publications: opportunity at "Advances in X-ray Imaging" in RAL (13 June); NPL X-ray User Conference (TBA July), !ToScA (6-8 September)
Link to RSE at Manchester and community:
http://rse.ac.uk/
Actions and Discussions
- Create Good-and-Bad data sets - @ ISIS (WK and DK) @ DLS (MB and DK) and in @ Lab (JJ, RC)
- Ring Artifact highlights: integrtae in savu version from DAWN (MB and SN)
- Missing Wedge algorithm (SN and PG and MB) - include published / paper outcomes
- Helical system (BL and SN) - include published / paper outcomes
- Very Noisy Data (??)
- Poster for Warwick Meeting with NPL standards - July 2017
- Poster / talk at ToScA - 6-8 September 2017
- Poster for Advances in X-ray Imaging, 12-13 June 2017 - convert to a leaflet for CCPi stand etc.
- ...
- Joint workshop with PETMR flagship (action MT / BL / KT) this should be an anual workshop topics to include;
- Baysian Methods,
- HW speed-ups and,
- SW framework links between flagships,
- ...
- PETMR (SIRF https://github.com/CCPPETMR/SIRF/wiki
) links and admin structure
- Possble link to Engin-X (ISIS) Joe Kelleher to help coordinate
- ...
- Integrating route? (with CCPi to Savu?) / Which platform? (Python vs Matlab) / Version control (CCPForge vs Github)
- Interaction with CCP PET/MR - annual workshop proposed
- Join RSE team at Manchester and in SCD: including courses
- ...
- Summer student request to CCPi "... laminography work I'd be interested to see taken a bit further than what I think will be possible during the semester - Work to be done: wrap up as tools for others to use, with documentation, link to Parmesh's acquisition code etc."
There was an extra meeting over a working lunch to discuss speakers for the Advances in X-ray Imaging workshop to be held on
13 June 2017; and the Fringe meeting to be held on
12 June 2017.
- There will be 8-9 speakers on the main workshop and Jenny Woods has a list to invite - please forward alternatives etc
- Course on savu as an introduction and advanced use - from Nicola Wadeson (MB)
- Training session on ASTRA and invite a speaker (CR, MB)
- Create a circular lapel badge - stateing "No FBP" (letters FBP with a red diagonal across it; c.f. to a no right turn road-sign)
- Topics for the Fringe event: "how to do the hacking" / "how to do Not-FBP" / "joint reconstruction an segmentation" /...
Follow up meetings:
Savu Conference Call, 11 May 2017: Participants: Nicola Wadeson, Mark Basham, Daniil Kazantsev, Sri Nagella, Philip Withers, Julia Behnsen, Jakob Jørgensen
Multichannel flagship project computational resources:
- All agree that two workstations to be acquired will be for smaller-than-production scale reconstruction, ie, proof of concept. Therefore specification not extremely critical. Also do not want too complicated setup, which might make methods developed specific to particular system, but a general, flexible setup, that other facilities may easily adopt.
- Larger scale production: Aim to use SCARF or Emerald(?) clusters. Should have access through IMAT(?).
- Savu uses MPI for parallelization. At Diamond each MPI process uses a separate GPU (exclusive access). GPU can be set to non-exclusive access mode, and Savu allows different processes to access same GPU. Hence many smaller GPUs not necessary (in practice having more than four in a workstation is difficult).
- Possibly best trade-off: Two mid-range GPUs instead of one high-end. Would probably allow same or more FLOPs with same budget, and allows to experiment with multi-GPU use, relevant for massively scaling up on SCARF/Emerald clusters with many GPUs.
- In the end, unclear whether decision was to still wait a bit for some Savu tests to happen in the next months before acquiring workstations, or given that workstations will be only for proof of concept, can proceed to acquire soon.
- Would be good to get ISIS-IMAT datasets soon to experiment with: Winfried Kochelmann has offered access during the current commissioning time for us to acquire data sets. Timing good before fall when users start coming.
Cone-beam in Savu:
- Currently only parallel-beam implemented. Data loading (incl metadata such as source-to-center, center-to-detector distances), internal representation, parallelization and reconstruction to be implemented. In part to be done by Savu team and by Daniil/Jakob (maybe wrapping ASTRA FDK), to be further coordinated during visit at Harwell.
Savu for laboratory micro-CT:
- 1st step: Get Savu up and running at existing workstation with single GPU in HMXIF, try running Savu's test data sets. Awaits new version of Savu released and hopefully installation guidelines.
- Possible 2nd step (only discussed by Julia and Jakob after meeting): Adapt for use with Zeiss Ultra nano-CT instrument data. This is parallel-beam monochromatic lab CT data. Requires importing into Savu of Zeiss data format, or of manually exported TIFF/BIN-stack plus metadata. Should be able to see benefits of using Savu functionality on this lab data with minimal effort.
- 3rd step: Adapt for use with cone-beam micro-CT data. Can use e.g. SophiaBeads or SparseBeads (Nikon format) datasets from Zenodo to test.
Questions:
- Are Python data readers for Zeiss and Nikon cone-beam data already available in CCPi core library? If yes, are or can they be easily used from within Savu?
- How (and when) to confirm availability to us of SCARF/Emerald clusters; how to initiate the scaling-up work?
Meeting at Working Grop and workshop: 12-13 June 2017; SN, JJ, DK, MT, EP
Presentation given:
Discussed items:
- Workstation - single larger machine based in Manchester to be purchased; action for quotes (DK, JJ)
- JADE (GPU cluster) login etc to be investigated MT
- Images and ppt for flagship - uploaded
- Datasets:
- ISIS/IMAT; Genovere meeting to be organised DK (aim for scan time before Xmas)
- Crab data set - XCT and Neutron tomography RG/IO
- Kate Dobson at Durham to be contacted JJ/MT
- Manchester hand-over for system in HMXIF - JJ to monitor and extract new test data
- Website update - tab on main site done but needs tidying up
- Move twiki information to drupal site (SN/MT)
- Submission for ToScA 2017 - done
- Leaflets/poster etc to be redone.
- update leaflet etc (SN to send around doc)
- Phantom: Andy King at soleil and psiche beamline. (MT to contact)
- VR graduate student is in RAL May-August: working with SN.
- Management meeting to be arranged for early September: then every three months.
--
MartinTurner - 24 Mar 2017