23rd July 2020

Meeting 23/07/2020

Agenda

Presentation: N/A

Chair: Yipeng

Minutes: Nina

Attendance:

Yipeng HY

Yunguan Fu YF

Ester Bonmati EB

Adria Casamitjana AC

Zac Baum ZB

Zhe Min ZM

Shaheer Saeed SS

Matt Clarkson MC
Qianye Yang QY

Nina Montana Brown NMB

Apologies:

Remi Delaunay RD

Updates and next tasks

Who?	Discussion Points	Next
Zac Baum	Let me know of any new/desired website updates
Ester Bonmati	Demo development: log error, epoch error, data in google drive (learn2reg), chmod, Labeled and unlabeled, same demo or different? Add labeled in name? Version? Demo readme application before instructions? Prerequisites Suggestion: demo_plot	Demo Medium MEC?
Adria Casamitjana	Quick questions: NCC kernel size Save niftis in prediction? Save pred_images in prediction?	Demo
Matt Clarkson
Remi Delaunay		Loader testing
Yunguan Fu	Prediction output(need a reference of a software to visualize nifti data) CLI tool, data format	CLI tool
Alex Grimwood		Finish loader testing and identify next tasks
Yipeng Hu	Demos progress Data, pre-trained models, instructions, Pictures of data loaders Any grouped loader demo Proofreading (all docs) - issue tag proofread Revised milestones Educational challenge JOSS paper (if and when)
Nina Montana Brown	Please review Edu challenge Lots of tests pending review - need pull requesters to update with master as this changes very often. Once this is done please ping your reviewer so they can resolve ASAP. We need to proofread the docs We need to provide sphinx docs	Submit edu challenge on Friday
Saeed Shaheer		Finish learn2reg t1 demo predict script Get started with ‘add new loader tutorial’
Qianye Yang		Longitudinal Demo MI Loss
Zhe Min

Open Questions

Minutes

Yipeng HY

Nina Montana Brown NMB

Yunguan Fu YF

Remi Delaunay RD

Ester Bonmati EB

Adria Casamitjana AC

Zac Baum ZB

Zhe Min ZM

Shaheer Saeed SS

Matt Clarkson MC

YH: Discussion points then potentially talk about pytest + codecov.

ZB: If people want more changes/ideas to put on the website please get in touch with Zac. Potential to add pictures? Raise tickets please.

EB: Questions on demo development:

1. Assigned a task to prepare a labelled and unlabelled examples. Two vs one demos? YH: Shaheer separated them, EB’s choice in the end. Add working demos now, and reformat for consistency later. YF: change the config yaml from true to false, change the loss so no label loss. YH: predict should be the same. YF: to ignore labels during training and then use them in inference is more challenging and will require some thought. We have many labels, do we need to worry about the rest.

2. EB: What is the current version of deepreg? YF: Tag with commit hash from master used for dev. YH: before first release, make sure that demos work with given tag. Need to come up with strat to maintain demos.

3. EB: demo instructions for README describes steps. Suggests swap application with instructions as the order is “instructions” and then “application”. YH: sounds sensible.

4. EB: add prerequisites settings for demos? Add a sentence on how to install/”requirement being DeepReg”.

5. EB: demos to be run as command line interface or as functions? YH: free for all. Depends on the scenario.

6. EB: add a “demo_plot” function/script as a requirement. Standardise the implementation visualisation for the results from DeepReg.

7. EB to meet with demo people (Shaheer, Yunguan, Nina) to discuss some demo errors. Quicker to discuss in person than to raise as ticket.

8. Medium page for MICCAI 2020 Edu Challenge? AG to volunteer help for Medium post for DeepReg. YH: argument for Medium, exposure, argument against: maybe not ready for traffic on repo right now.

AC: Post in the group about questions to make things quicker or ask the questions afterwards. YH: worth discussing some things.

1. Kernel size: how to specify best parameters in config file?

2. AC: in label_loss function -> specific argument for image parameter. Which is passed in via the config file. Add flexibility: parameters like gaussian kernels. YH argues it may not be that important, but we can set a default value in the config file.

3. Resolve test PR.

YF:

1. refactoring predict file. Imagines image outputs: all the existing outputs and warped (images, labels). Ssd, dice, tre for label. Mean, std, median across batch, index…

Images are saved as pngs for the moment. Contain as h5? Or nifti? Save ddf? Save image? AC: argues if you have image input and output can compare. So would be useful to save this instead of just the ddf. If there are several subjects, these may be “overwriting each other” - only saves the last one. The code doesn’t save the files into separate folders.

YH: sounds like a bug -> raise ticket suggested. Review next week. Which softwares can visualise nifti? YH: Slicer.

2. CLI: two input nifty, one output nifti.

AG:

1. Does everyone have enough data for demos? 3D Phantom us data. YH: working on US CV.

2. Finish testing, get on to reading/proofreading/user testing.

YH:

1. Review demo progress. Need to review timeline for demo delivery - complicated by development of repo. Needs a bit more time. Asks demo writers to use best judgement for speedy delivery. Downloading and preparing the data is not trivial. Do not assume things by user. Add a pre-trained model. YF: write a bash script to check that it works without erroring on PT towers. NMB: how do we know if the demos work? They may not error but that doesn’t mean it will be significant. YH: discuss demo testing - ensuring that it works - at a later stage.

2. NMB: on line of testing demos, SS to send pre-trained model for lun reg demo to NMB to ensure the demo works and to add to edu challenge.

3. YH: illustration of the loaders to simplify the explanation of how they work. 3 illustration - get in touch.

4. YH: grouped loader demos. Who is in charge of this?

5. YH: proofreading. AG spotted a couple of issues. Stick in a single issue? YH: yes. EB: suggestion - ticket for typos, point from which file. Do not open a single ticket for one error.

6. YH: JOSS - still a good idea? Should we sent to a different journal? Once there is consensus, find a target date for submission. EB: Computer methods and programs in biomedicine? YH: always ask for scientific novelty in this journal which we currently don’t have. That’s why we thought of JOSS first. Any comments or feedback appreciated?

NMB:

1. Please address your assigned PR so we can resolve tests.

2. Someone needs to sphinx-ify docs. 3-4h work. A couple of other 1-2h tickets - please help out!

3. We are submitting an edu challenge for MICCAI 2020 - everyone author as we are an OS project. If you have a problem please get in touch please, and review.