15th April 2021

Meeting 18/02/2020

Attendance:

Alex Grimwood AG
Zac Baum ZB
Ester Bonmati EB
Adria Casamitjana AC Yipeng Hu YH
Lu Sicong LS
Min Zhe MZ
Nina Montana Brown NB
Shaheer Saeed SS Kyriaki-Alkisti Stavropoulou AS
Wang Zhiyuan WZ

Apologies:

Yunguan Fu YF
Bryan Ko BK
Matt Clarkson MC
Qianye Yang QY
Mark Pinnock MP
Remi Delauny RD
Stefano Blumberg SB
Yiwen Li

Agenda:

Welcome to MSc students Sicong and Zhiyuan!
Review of tickets and assignments

Ticket Review:

#162 - YF and YH test missing data management

#423 - NB review docs: user feedback indicates autogenerated docs for API are confusing due to missing descriptions

• #423 is larger than just API: datasets, loader etc. all need refactoring (with people assigned to each task)
• API issue should be its own ticket
• YH: redistribute and assign - possibly to YH, AG & EB
• TBC: for R1/RC1

#468 - YF support mixed precision calcs

#473 - ZB 3D slicer integration: needs to be shelved until after RC1

#475 - ZB point-set registration: feature will not be integrated until RC2

• AS, NB and ZB all agree to contribute
• ZB to organise meeting offline

#476 - Automated hyperparameter tuning

• YH ticket requires more discussion around what to implement
• YH, SS will discuss in ticket thread. Encourages others' opinions too.

#536 - New demo for globalnet

• AC happy to implement: goal is for clear evidence of alignment in the output

#564 - Update visualisation tool docs

• SS happy to continue

#586 - Dataset build without generator: to be discussed offline

#608 - Release to docker hub: to be discussed offline

#610 - complete pylinting: NB and others, linked to #423

• requires dividing and assigning to multiple people
• NB: requires manual correction to attain PEP8 compatibility is at odds with precommit code

#619 - resampler improvements: to be discussed offline

#676 - raise errors instead of asserts: ensure coding best practice

• YH to seek volunteers to review and correct current codebase

#678 - MobileNetV3 for dense prediction

• LS assigned to ticket: add a new architecture and to assess improvements to training time/accuracy
• YH to discuss offline

#680 - improve model structure plot: open to any volunteers

#700 - automated docs for argparse: discuss offline

#703 - velocity-predicting option demo: WZ

• WZ assigned, YH to discuss offline and recommends unpaired CT data

#709 - add support for weights and biases

• possible duplication with Neptune.ai (#710)

#716 - Benchmarking of key functions: open to volunteers, let YH know offline

• ticket can be broken down into specific tasks

#723 - Standardising Gaussian kernel definitions: YH anyone welcome to try

#744 - iterative registration:

• YH separating metrics and losses are dealt with differently in TF/Keras, making implementation of this ticket challenging