DIALS Meeting 2021-03-31
CZI Image Viewer Proposal
- David: Successfully put in application for Image Viewer project to the CZI
- Strong support from MX community in general, so might be general support for this even if funding application is unsuccessful
- 20th April is where we find out results of this round of consideration
- Applied describing project as “Web First”
Numpy-spotfinding Hackathon
Results of intial hackathon working towards implementing the DIALS spotfinding algorithm in pure python summarized by Elena & Ben:
- Started looking at various image handling tools - OpenCV quite wedded to Integer Images (floats not heavily supported) and common format bit-depths
- Noemi/RG started on Numpy Approach
- Elena: We started with normal loop, moved to numpy, tried to reimplement in Python the algorithm used in DIALS. Then tried to apply Numba to python code, saw improvement in speeds - hybrid code using both numpy/numba was best performance we saw - about half the speed of the standard DIALS implemention - ran out of time
- Implemented full DIALS thresholding algorithm - so SAT and thresholding calculation - matching the non-extended dispersion technighe. Validated that numbers match
- Ben feels that the approach of Numba is preferable, and seems like a good approach to move forwards with - good outcome for a one-day sprint
- Most efficient SAT calculation we tried was a Numba-compiled simple for loop
- Script concluding the days efforts: https://github.com/dials/dials_scratch/commit/bd300294bf3c7f4d256be983eab64307b4dd665c
- Short discussion on whether GPU was explored, no - not this time
- Nick pointed out that many of these questions about performance and platform (GPU etc) likely fall into his responsibility area.
- Future path: Put a notebook around the script we made so people can read - and then make another hackathon session in future
- When to schedule? Discussed dates, April seems pretty unlikely to get all interested parties at once. Discussed possibility of 14th May
- Action Item: Elena to write up results from session
- When to schedule? Discussed dates, April seems pretty unlikely to get all interested parties at once. Discussed possibility of 14th May
Profile fitting zero-intensity problem
- JBE: #1640 - Profile Fitting is showing up with zero weights after change in #1297 e.g.
- Before #1297 we didn’t allow any invalid pixels, now we allow all invalid pixels - in these cases profile fitting is meaningless
- Potentially severe bug - Might be worth pushing into release fast
- DW: This is sensible approach and same approach as XDS - definitely a bug
- Decision on fraction to accept needs to be made - 0.75 is same as XDS, JBE thinks needs to be at least 0.5, 0.75 probably a safe value.
- Action: JBE to continue looking at this today
DIALS/Neutron-diffraction Project
ND asked about Dials-Neutron Diffraction project, and a general update about how that was going on
- DW: Is going ahead as part of a pretty focussed work package - is not really overlapping with dials currently and will move closer in the future
- Saw reciprocal lattice with results from Neutron TOF and looked impressive
- Geometry viewer useful! Nice to see full sphere of detector surfaces, helped identify couple of wrongly placed panels
- Graeme wanted results presented to group in future, need to organise