oeway commented 4 years ago

Hi,

The BioImage.IO team will meet every Wednesday 4PM CET, check the next meeting in your time zone.

Meeting Format

As discussed during in the AI4Life general assembly meeting (Oct. 10th, 2023), we will make the meeting more structured. For the first 15 minutes, we will take turns and having one presentation every week from one of the AI4Life partners, or guest speakers from community partners, developer etc.

The aim of the first 15-minute presentation is to provide community-friendly (especial for users and developers outside the AI4Life consortium) updates on new features, changes, announcements etc.

And the later part, we will have regular internal discussions or other relevant topics. Everyone is welcome to join the meeting via zoom: https://kth-se.zoom.us/j/65777152331, feel free to add your name and topic to the meeting agenda.

oeway commented 4 years ago

June 26th, 2020

Attendees

Name	affiliation	username
Wei	KTH	@oeway
Esti	UC3M	@esgomezm
Hao	KTH	@haoxusci
Arrate	UC3M	@ArrateMunoz
Anna	EMBL	@akreshuk
Fynn	EMBL	@FynnBe
Maxim	EMBL	@m-novikov

Agenda + Minutes

All: general intro and your plans for the zoo in the near future
[name=Wei]:
- updates on the new bioimage.io website
  - rebuild the website with a build system and CI
  - New UI design powered by Bulma
  - Support models, applications, notebooks, datasets
  - Support badges
  - Share the site template with other projects: covid19ai.io, shareloc.xyz etc.
  - Building backend (imjoy-ai-engine) for supporting uploading datasets and online inference service
[name=Anna] Technical question: single model with multiple weights, trained in different ways
[name=Anna] Other potential collaborators: ZeroCostDL4Mic
Discussion
- How to support linking datasets with model weights and describe the training
- Extend the model description spec, also for describing datasets, notebooks, applications

Note

How to group different cards
- if only weights changed == same card
- If postprocessing changed == different card

oeway commented 3 years ago

July 10th, 2020

Attendees

Name	affiliation	username
Wei	KTH	@oeway
Esti	UC3M	@esgomezm
Daniel	EPFL	@dasv74
Anna	EMBL	@akreshuk
Carlos	UC3M	@carlosuc3m
Maksim Novikov	EMBL	@m-novikov
Ricardo Henriques	IGC/UCL	@paxcalpt

Agenda + Minutes

[name=Wei]
- Updates for the website
  - resource description file
  - support links between items (apps to notebooks, dataset etc.)
  - support pagination
  - add ZeroCostDL4Mic and HPA as a community partner
  - comment feature powered by github
  - community partner
    - what's the criterion and procedure to join as a community partner?
      - Should we allow their own model zoo? (Think image.sc)
      - Open source? comercial partners?
        
        Anna: accessible dataset, source of the model
        
        Esti: published model without close dataset
        
        Wei: sensitive medical data
        
        Ricardo & Wei: Use badges to indicate and encourage
        
        conclusion: give recommendation and show badges
      - Allow at least one admin of bioimage.io to access the repo?
        
        conclusion: agreed
    - what kind of role and responsibility does the community partner have?
      - Contribute and maintain models/applications/notebooks
      - Accept third-party contribution if relevant? Say ilastik will in charge of all PRs for the ilastik models?
        
        Anna: for long term, we should see more interoperatable models and that should go to BioImage.IO
        
        for dedicated models that works for only one community partner, contribute to the partner instead of BioImage.IO
[name=Anna]
- What examples do we need to add to make it easy for people to understand how to run models?
- We should add a general "About" page describing what and why
  - Wei: Will do, this already been added for the community partners
[name=Daniel]
- Demo of Model manager in DeepImageJ
- Suggestions by Wei:
  - Add bug report button to link the same github issue as on BioImage.IO
  - Show linked items
[name=Esti]
- Good practices uploading models; i.e. authorship & responsabilities
  - Anna & Ricardo: cite and acknowledge properly, typically it's fine to upload models that are derived from others' work
  - Wei & Ricardo: follow the license, most license will include terms about attribution
Discussion
- What kind of users we are targeting with BioImage.IO
- What's scope and focus of the bioimage.io platform? Should we only provide models, or including also the downstream/upstream utility tools for building workflow. The layout of the website.
- How to upload a model compatible with different software? Same card / id besides proper reference?
  - Wei: use CI to automatically test the software compatibility, using BioEngine api
  - Anna: These tests should be done automatically
Suggestions
- Daniel: Allow users to post images to the github issue for each model
- Esti: support linking to image.sc
- Ricardo: Network view for navigation

esgomezm commented 3 years ago

July 24th, 2020

Attendees

Name	affiliation	username
Esti	UC3M	@esgomezm
Maksim Novikov	EMBL	@m-novikov
Arrate	UC3M	@amunoz
Constantin	EMBL	@constantinpape
Anna	EMBL	@akreshuk

Minutes

Discussion

What’s the distinction between models and notebooks -- DeepImageJ: Notebooks and Models are actually separate -- Notebook is used for training the model -- models are then only the train -- more general, how do we link models, notebooks (e.g. used to train the model) and datasets (e.g. where model was trained on)
datasets: -- how do we reflect “small” changes to the data-sets, e.g. fixing issues in the ground-truth -- public datasets can be “easily” added -- what about the private datasets? -- change the name? maybe "working data", "training data"
easier way to explore the models apart from search, ideas: -- “structured tag cloud”: selecting multiple tags to find a model that fits the “segmentation”, “electron-microscopy”, “membranes”, …
Discussion about the CZI Open Software call https://chanzuckerberg.com/rfa/essential-open-source-software-for-science/ DeepImageJ is considering to apply; question is if/how to integrate the BioImage.io in there. Including Demo resources. Ilastik is also applying for the call, but for a different purpose? Are other community partners applying? If so, it would be nice if all could mention the model zoo. bioimage.io is not yet mature enough to apply on its own
once the website is more stable, we could start to write user facing docs a good example would be integration of the different community partners, e.g. train model in a ZeroCost notebook and run it with another community partner software
started discussion on interoperability between tools by community partners; big issue is (still :)) the different pre and post-processing steps that are implemented in a language specific manner. For now, porting between tools will probably be done manual by a developer of that tool.
When to change bioimage.io as landing page / reference model page for all community partners (e.g. for DeepImageJ, CSBDeep, ZeroCost home pages)? Then we could announce on image.sc etc. E.g. First of November. Discuss with Wei. Need to discuss with other community partners.
Citation discussion; how many citations to put in

Suggestions for the website

How do we make the website more intuitive for users?

Name-changes:

Change the name "Datasets" to "Training Data"; this should be clear for users.
Change the name "Models" to "Preconfigured Models"
Change the name "Notebooks" to "Code"

Could also go with more actionable names:

"Models" to "Find Models"
"Applications" to "Preview" or "Try on your data"?
"Notebook" to "Find Code"
"Datasets" to "Find Data"
Get rid of the "Application" button and just provide "Run" or "Demo" button on the model windows

Icons on the Model and Notebook Cards are different for the each Model / Network. This is confusing.

oeway commented 3 years ago

BioImage.IO Meeting Minutes

August 21st, 2020

Attendees

Name	affiliation	username
Wei Ouyang	KTH	@oeway
Emma Lundberg	KTH	emmalundberg
Ricardo Henriques	UCL / IGC	@paxcalpt
Guillaume Jacquemet	ÅAU/UTU	@guijacquemet
Anna	EMBL	@akreshuk

Minutes

planning for the release of Bioimage.IO
- release date: November 1st
- features to offer by then:
  - display plain model cards for download
  - support interoperatability? @emma high priority
    - a list of useful/popular model, make sure it compatible with a few software. (show complementary with software rather than competition)
      - pytorch model (segmentation): ilastik, imjoy, zero
      - tensorflow model (denoising): zero, fiji, deepimagej, imjoy
      - classification: different types
      - starting from a training dataset? task?
  - use outline color encoding for types @Anna
  - show text description for tags
  - online inference (single model, save it for the release)
  - model searching application (lower priority)
  - integration with colab, binder
  - community parties
    - guidelines for the community parter
    - short recommendation from all the community partner
    - feature models, sorting, popularity, model searching (lower priority)
  - Group the software by three types of tasks: 1) preprocessing/ annotation 2) training / transfer learning 3) prediction. Click a model, show a list of software grouped by tasks types
  - Improve 2h rendering interval @Ricardo
  - parser script should validate the link, valide the fields, cover image size, add badges

oeway commented 3 years ago

September 4th, 2020

Attendees

Name	affiliation	username
Wei Ouyang	KTH	@oeway
Constantin Pape	EMBL	@constantinpape
Anna Kreshuk	EMBL	@akreshuk
Maksim Novikov	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Ricardo Henriques	IGC/UCL	@henriqueslab
Estibaliz Gómez	UC3M	@esgomezm

Agenda

Wei
- Outline color encoding
- Zarr dataset and Vizarr
- ImageJ.JS
- discussion
  - Run DeepImageJ models
  - Model conversion service
    - BioEngine Application running on Binder
    - Tensorflow to Tensorflow.JS
    - PyTorch to ONNX to Tensorflow?
  - RPC link between the website and software
    - From the website: Send model from the website to Ilastik/Fiji etc.
    - From the software: launch model browser
    - Example PycroManager
- Ilastik
- Specification of training, Application specific?
- Website header too large?

Minutes

Add underline color LUT
What models to use as show case
- 2D Unet / Nuclei Broad
- 3D Unet
Token (url) for linking model across software
Roadmap general issue
Training:
- descriptive training procedure is too ambitious
- instead: partners can support their own training procedures; training APIs can be shared
- example: ilastik will have a training API that can also be exposed through imjoy RPC so it could be used from the browser

oeway commented 3 years ago

September 18th, 2020

Attendees

Name	affiliation	username
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Carlos García	UC3M	@carlosuc3m
Estibaliz Gómez	UC3M	@esgomezm
Novikov Maksim	EMBL	@m-novikov
Florian Jug	CSBD + HT	@fjug
Fynn Beuttenmüller	EMBL	@fynnbe
Maksim Novikov	EMBL	@m-novikov

Agenda

Roadmap general issue
- Model runners
- Improve Landing
- Documentation
- animal icon?
- Run button for each model
- (Simple) models which can be run in multiple applications. E.g. train a model in a notebook in ZeroCost, run in both DeepImageJ and CSBDeep (or ilastik) (note: deal with output error produced by different framework)
  - random thoughts from the ilastik side: https://docs.google.com/document/d/1FUqEzEuwEi8Z_XULl4MgisDl-eX0OV6QKBQwxEcBQbk/edit?usp=sharing
Cross-compatibility of models - how much can we guarantee for a model following the specs?
Generic model runner (two runners, PyTorch and Tensorflow) to have the same as HPA model runner for all models?
What should really be in the "Applications" part?
New applications: CellPose, ImageJ.JS, CellPose-ImageJ.

oeway commented 3 years ago

October 2nd, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Florian Jug	CSBD + HT	@fjug
Dominik Kutra	EMBL	@k-dominik
Jan Funke	Janelia	@funkey
Carlos García	UC3M	@carlosuc3m
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz

Agenda

Florian
- Pitch the model zoo project
Wei
- Updates on the bioimage.io website
- BioEngine API for model runners
```
 api.registerService({
    "type": "model-runner",
    "init": init, // initialize the runner
    "predict": predict, // perform prediction
    "fit": fit // perform training (optional)
})
```
  Alternatively, we can use scikit-learn style with fit and transform function.
  - ilastik
  - deepimagej
    - support via pyimagej
    - support via imagej.js
- Demo for a new model: Learn2Sofie
- Jan
  - Introduction about the live supervised learning platform at Janelia
- Anna and Wei
  - plan for the November release, implementation of the model runners.

oeway commented 3 years ago

October 30th, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Tom Burke	CSBD	@tomburke-rse
Florian Jug	CSBD + HT	@fjug
Deborah Schmidt	CSBD	@frauzufall
Arrate Muñoz	UC3M	@arratemunoz
Anna Kreshuk	EMBL	@akreshuk
Novikov Maksim	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Dominik Kutra	EMBL	@k-dominik
Estibaliz Gómez	UC3M	@esgomezm

Agenda

Discuss the test_inputs and test_outputs format: will be feeded as a list and use npy as the only formats, this will be specifically used for CI testing, but combined with converters, for the users, we can use another field, e.g. test_samples
Change normalization to preprocessing, and agreed on several types of preprocessing functions
The model file can contain multiple formats of the same weights, the weights is a dictionary which uses a unique format string as the key

See also in the project: https://github.com/orgs/bioimage-io/projects/1

oeway commented 3 years ago

November 13th, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Tom Burke	CSBD	@tomburke-rse
Florian Jug	CSBD + HT	@fjug
Deborah Schmidt	CSBD	@frauzufall
Arrate Muñoz	UC3M	@arratemunoz
Anna Kreshuk	EMBL	@akreshuk
Novikov Maksim	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Dominik Kutra	EMBL	@k-dominik
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Carlos García	UC3M	@carlosuc3m

Agenda

0.3.0 project: https://github.com/orgs/bioimage-io/projects/1
- Authors in weights vs. authors in model
  - 3 parties to give credit to:
    - Who is the creator of the bioimage.io.model.zip? The creator in the yaml points that out. The top-level description describes the training setup, purpose, etc.
    - Who is the creator of the type of model? The existing list of citations is good enough!
    - Who trained the linked weights? If not the same as the creator of the model, names can be given here. Description field is superfluous and needs to be removed here!
- Weights-formats we support:
  - _tensorflow_saved_modelbundle: Zip the weights and the folder with the variables and link to this zip from the yaml-file. (File name: *.zip)
  - _kerashdf5:
  - _tensorflowjs: a json file + binary weights files
  - _pytorch_statedict:
  - _pytorchscript:
- Naming convention for the model bundle: *.bioimage.io.model.zip
- show a table with features (weights format, pre/post processing types) that are implemented by consumers

oeway commented 3 years ago

November 20th, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Florian Jug	CSBD + HT	@fjug
Deborah Schmidt	CSBD	@frauzufall
Arrate Muñoz	UC3M	@arratemunoz
Anna Kreshuk	EMBL	@akreshuk
Novikov Maksim	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Dominik Kutra	EMBL	@k-dominik
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Carlos García	UC3M	@carlosuc3m
Ricardo Henriques	IGC & UCL	@henriqueslab

Project: https://github.com/orgs/bioimage-io/projects/1

Agenda

Suggested Agenda Points from Dresden
- I2K - 0.3.0 releases? Our own dialects? What are our goals?
  - Let's make it 0.3.0 and not any of our dialects!
- More generally: when and how would 0.3.0 count as 'released'? How can we ensure that the released state leaves no margin for interpretation? (Example: how would I find out what parts are optional?)
  - For now, the readme in the repo is the full spec. We will soon release on GitHub the 0.3.0 version and that's it!
- (Minor?) Pre/-Postprocessing: could axis-concept be more ubiquitously supported? (Example: Pre::zero_mean_unit_variance vs. Post::scale_linear)
  - We will add axis to scale_linear (and maybe more?) after I2K and freeze 0.3.0 in its current form.
- Input/Output types or class? Example: if you have a model that gets/produces an image and a table of datapoints. How would I know what is what? Where would I specify this best? (Wei: Hints for rendering of inputs/outputs tensors for UI, proposal)
  - Let's start a 0.3.1 branch and get going. Plenum discussion after I2K in December...
How do we distinguish between the models that do not need pre/post-processing and the ones that cannot be described with the defined steps?
- DD thinks: In that case we use the 'source'-tag... (and we do that already for N2V).
Metadata of the images and numpy arrays?

Decided:

blow up if config key contains anything unfamiliar
custom pre / postprocessing should also go into the config

TODO befor releasing 0.3.0:

finalize weight formats and pre/postprocessing here: https://github.com/bioimage-io/configuration/pull/43
Fynn will go over the configuration Readme and mark everything that is optional in a consistent manner

oeway commented 3 years ago

November 27th, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Deborah Schmidt	CSBD	@frauzufall
Tom Burke	CSBD	@tomburke-rse
Arrate Muñoz	UC3M	@arratemunoz
Fynn Beuttenmüller	EMBL	@fynnbe
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Novikov Maksim	EMBL	@m-novikov
Carlos García	UC3M	@carlosuc3m
Daniel Sage	EPFL	@dasv74

Project: https://github.com/orgs/bioimage-io/projects/1

Agenda

What's left to do before releasing 0.3.0?

Outcome

change optional from [field] to field [optional] https://github.com/bioimage-io/configuration/issues/53
instead of changing failing for unknown fields in config
- allow for arbitrary fields in config
- have a execution_model that specifies how to run the model. this also has optional kwargs
- custom pre/post-processing for deepImageJ could go into MacroExecutionModel
- https://github.com/bioimage-io/configuration/pull/51
- meta image information, e.g. physical image size, probably can't solve in 0.3.0 probably

More user-friendly website

add a introductory tour with several steps (easy option would be a tutorial video)

Dev documentation

improve md based spec documentation

Run torchscript (e.g. models from deepImageJ) in ilastik, (after ilastik can parse 0.3.0)

Autogenerating the spec documents from yamls or spec parser

https://github.com/bioimage-io/configuration/issues/52

oeway commented 3 years ago

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Tom Burke	CSBD	@tomburke-rse
Arrate Muñoz	UC3M	@arratemunoz
Fynn Beuttenmüller	EMBL	@fynnbe
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Novikov Maksim	EMBL	@m-novikov
Daniel Sage	EPFL	@dasv74
Florian Jug	CSBD + HT	@fjug

Agenda

Definition of model ID for the parent field: use the URL or path to the yaml file, change the key name to uri. @constantinpape
Update to configuration: https://github.com/bioimage-io/configuration/pull/59
Data type and data range: https://github.com/bioimage-io/configuration/issues/55#issuecomment-743054525 We agreed on the following data flow:

oeway commented 3 years ago

December 18th, 2020

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Florian Jug	CSBD + HT	@fjug
Carlos García	UC3M	@carlosuc3m

Project: https://github.com/orgs/bioimage-io/projects/1

Agenda

We have a very short meeting today, but still quite fruitful for finalising the last pieces left for v0.3.0!

Update to configuration: https://github.com/bioimage-io/configuration/pull/59 Update to update: https://github.com/bioimage-io/configuration/pull/61

Data type and data range: https://github.com/bioimage-io/configuration/issues/55#issuecomment-743054525

https://github.com/bioimage-io/configuration/pull/61
epsilon for zero mean, etc default: 10e-7 (used by tensorflow)

oeway commented 3 years ago

January 15th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Daniel Sage	EPFL	@dasv74
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Ricardo Henriques	IGC & UCL	@HenriquesLab
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Jan Philipp Albrecht	MDC	@jpalbrecht
Deborah Schmidt	MDC	@frauzufall
Carlos García	UC3M	@carlosuc3m
Anna Kreshuk	EMBL	@akreshuk
Fynn Beuttenmüller	EMBL	@fynnbe
Maksim Novikov	EMBL	@m-novikov

Agenda

News: BioImage Model Zoo featured on Nature news https://www.nature.com/articles/d41586-021-00023-0
Documentation
- User Guide
  - Overview
    - Add the cross-compatibility table
  - Tutorials
    - including videos
  - How-tos Guides
    - How to choose a model for my data
    - How to run this model in community partners' tools
    - How to search by tags
  - Explanations
    - The risk of model mismatch
- Model Developers
  - Tutorial
  - How-tos Guides
    - How to make a model package
- Application Developer
  - Tutorial
  - How-tos Guides
    - How to become a community partner
    - How to build a BioEngine App
  - Explainations
    - Responsibilites for community partners
  - API reference
Vision
- Advanced AI models in one-click
Roadmap
- Runners support 0.3.0; tests for runners and models; CI
- Each model should be associated with at least one model runners
- Support type info for guide the rendering of inputs/outputs tensors
  - https://github.com/bioimage-io/configuration/issues/12
  - Next iteration of spec; experiments before possible
- A new tab for hosting scripts? (e.g. Fiji/ImageJ macro, similar to notebooks)
- RPC interface
  - Support making RPC calls to local software
  - Connect model API servers for support cloud based training and inference (e.g.: cellpose)

Timeline / meetings
- 3 weeks from now (Feb. 5th): compatibility hackathon
  - Ilastik, DeepImageJ, ImJoy, Zero?, Fiji
- 5 weeks from now (Feb. 23rd): documentation polish
- end of February: guinea pig users and contributors experiment
- early March: release!

oeway commented 3 years ago

January 22nd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Jan Philipp Albrecht	MDC	@jpalbrecht
Fynn Beuttenmüller	EMBL	@fynnbe
Maksim Novikov	EMBL	@m-novikov

upcoming Hackathon project: https://github.com/orgs/bioimage-io/projects/2

Agenda

TODOs before the hackathon
- Update most model config files to 0.3.0
- Model validation using https://github.com/bioimage-io/python-bioimage-io/
- Model runner for CI:
  - tiktorch
  - deepimagej
  - fiji
Hackathon planning Main goal: convert to different formats (e.g.: ONNX)
- Currently supported model formats:
  - tiktorch (pytorch, torch script, onnx)
  - deepimagej (torch script, tensorflow saved model, ~~keras h5~~)
  - Fiji (tensorflow saved model)
  - ImJoy (support all types of models with different plugins, including onnx, onnx.js, tensorflow.js)
- Model cross-compatibility aims:
  - tiktorch <-> deepimagej (torch script)
  - tiktorch -> ImJoy (onnx)
  - ImJoy -> deepimagej (tensorflow saved model) Idea: all the consumer convert their model into onnx, and use it as a neutral model representation, which can be further converted into other formats.
Date: Feb. 5th, 9am - 6pm

left for future agenda

where do the CI tests run that test models? https://github.com/bioimage-io/bioimage.io/issues/85

oeway commented 3 years ago

January 29th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Jan Philipp Albrecht	MDC	@jpalbrecht
Fynn Beuttenmüller	EMBL	@fynnbe
Maksim Novikov	EMBL	@m-novikov
Constantin Pape	EMBL	@constantinpape
Carlos García	UC3M	@carlosuc3m

Project: https://github.com/orgs/bioimage-io/projects/1

upcoming Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Generate CI error report for Models

The idea is to provide a link to the CI status as an additional field in the manifest.bioimage.io.yaml file: {"ci_status": "url"}

Then the link point to a json file contains CI status for the models:

{
    "name": "Ilastik CI",
    "timestamp": "2021-xsssxxx",
    "status": "",
    "error": "",
    "models": {
        "<Model ID>": {"status": "pass", "error": ""},

    }
}

The <Model ID> is the same as here: https://github.com/deepimagej/models/blob/master/manifest.bioimage.io.yaml#L38

This json file can be uploaded to a separate branch of the model repo, and this is an example which contains Github Actions configurations for uploading artefacts to another branch: https://github.com/bioimage-io/bioimage-io-models/blob/master/.github/workflows/compile-manifest.yml

The CI status json file will be loaded from the website and rendered as badges for model status.

oeway commented 3 years ago

BioImage.IO Hackathon on Model Cross-compatibility

Feburary 5th, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Joran Deschamps	MPI-CBG	@jdeschamps
Maksim Novikov	EMBL	@m-novikov
Romain F Laine	MRC-LMCB, UCL	@Romain-Laine
Guillaume Jacquemet	ÅAU	@guijacquemet
Carlos García	UC3M	@carlosuc3m
Florian Jug (dropped by)	CSBD + MPI-CBG	@fjug

Project: https://github.com/orgs/bioimage-io/projects/1

upcoming Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Brainstorming on what model/approach to work with Please write the potential model/apporach you can potentially work on here:
- Pytorch - TorchScript compatibility
- ONNX weight format
- bundling TensorFlow models directly from python (under development https://github.com/deepimagej/pydeepimagej/tree/bioimage-yaml)
- linking material from other community partners in manifest.yaml (e.g. ZeroCostDL4Mic notebooks with bundled models)
- update CSBDeep python and java parts to 0.3.0 + model runner integration into fiji
- List the versions we are compatible with. e.g. Fiji supports TF until 1.15
- Import ONNX models to ZeroCostDL4Mic for transfer learning and export models using ONXX for using models in other tools.
- numpy reader for ImageJ/Fiji(?)

@Wei - ImJoy

Loading DeepImageJ/FIji models via PyImageJ
ONNX.js models
Tensorflow.js models
tried ONNX.js multiple version: didn't work with the Unet model

@constantinpape - Ilastik

onnx models
torchscript models

@Esti - DeepImageJ

tensorflow saved_model.pb
Pytorch - torchscript

@Tom - Fiji/CSBDeep

Tensorflow1 saved models

@Joran

ONNX version of imagej-tensorflow

@m-novikov

running tensorflow models (e.g. U-Net pancreatic cell segmentation)
1. Work on the models
model with onnx and torchscript weights: https://oc.embl.de/index.php/s/0gHEyx2HJvVVlBt
Zero: import onnx models
Ilastik:
- Export onnx models
- ONNX runtime
ONNX runtime
DJL: can import onnx model https://djl.ai/
Tool for network visualization: https://github.com/lutzroeder/netron
1. Debriefing
- Demo for running ONNX runtime
  - TODO: make docs for loading yaml and do inference using ONNX runtime
- Changes in the spec 0.3.1 to support multiple version of the same weight format Proposal:
  - onnx:
  - sha256: 5bf14c4e65e8601a source: ./weights_opset11.onnx opset_version: 11 parent:
  - sha256: 5bf14c4e65e86 source: ./weights_opset12.onnx opset_version: 12 test_outputs: ["test_output_onnx_opset_12.npy"] parent:
- - Dimension ordering for different formats? (e.g. Pytorch vs Tensorflow) Proposal: Add transpose_axis for different dimension ordering
- ONNX version
- Operator Kernels https://github.com/microsoft/onnxruntime/blob/master/docs/OperatorKernels.md
- ONNX Versioning https://github.com/microsoft/onnxruntime/blob/master/docs/Versioning.md

Framework	ONNX opset_version	comment
ONNX runtime (Python)
ONNX runtime (Java)	12	version: 1.6.0
ONNX.js	9
onnx2keras
keras2onnx

Support Unet in different consumers

Consumers	Works?	comment
pytorch -> onnx(ilastik)	yes	except the output border
pytorch -> torchscript(deepimagej)	yes	except the output border (it works without tiling)
pytorch -> onnx(ONNX.js)	No	blocked
onnx -> keras (zero)	No	blocked
keras -> onnx (zero)	?
Fiji	?

oeway commented 3 years ago

Feburary 12th, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Joran Deschamps	MPI-CBG	@jdeschamps
Maksim Novikov	EMBL	@m-novikov

Project: https://github.com/orgs/bioimage-io/projects/1

Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Checking model spec (0.3.0)
- notebooks currently not compatible
- old discussion on json format for errors https://github.com/bioimage-io/bioimage.io/issues/28#issuecomment-769871722
- Post issues to https://github.com/bioimage-io/python-bioimage-io/issues to resolve errors
- Checker instructions here https://github.com/bioimage-io/python-bioimage-io
hackathon follow-up
- model with different weights https://oc.embl.de/index.php/s/mKXYRcmjWDTiYtY (need to fix the checker to support all weight formats)
- add tag cross-compatible and update the model
- json schema https://github.com/bioimage-io/python-bioimage-io/blob/master/json_specs/model_spec.json
- changing the weight format: https://github.com/bioimage-io/configuration/issues/63 (goal for 0.4.0)

oeway commented 3 years ago

Feburary 19th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Anna Kreshuk	EMBL	@akreshuk
Maksim Novikov	EMBL	@m-novikov
Jan Philipp Albrecht	MDC	@jpalbrecht

Project: https://github.com/orgs/bioimage-io/projects/1

Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Plan for the upcoming hackathon on documentation
- Discuss the priority for the release
- How to contribute models
- Model Documentation Templates with Runner
- Focus more on the developers
Website updates:
- Running pybio spec verification in the root model repo and add error information if a model fails
- Displaying a spec badge with passing and failing status for all the models, please fix all the failing models!
- Resource Item ID rule change: they will be all prefixed with the collection/partner ID, e.g.: deepimagej/WidefieldFitcSuperResolution
Upcoming changes:
- Add a new tab about scripts
- Improve the BioEngine, Model runner interface change
  - Support Ilastik/Tiktorch model runner
  - Support DeepImageJ.JS

oeway commented 3 years ago

BioImage.IO Hackathon on Documentation

Feburary 23rd, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Anna Kreshuk	EMBL	@akreshuk
Fynn Beuttenmüller	EMBL	@fynnbe

Project: https://github.com/orgs/bioimage-io/projects/1

Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Hackathon Documentation (https://bioimage.io/docs/)
- Discuss the priority for the release
- How to contribute models
- Model Documentation Templates with Runner
- Focus more on the developers
- Roadmap

Proposals

1.- What is the bioimage model zoo? It is more than advance AI models in one click, no?

"About" button. It points to https://github.com/bioimage-io/bioimage-io-models.
I would add a simple graphical schema about how to contribute, inputs and outputs of the bioimage.io to get a general idea about all the environment (google drawings?)
In the contribution text, it says: "You are welcome to submit your models, datasest, applicattionionss and Jupyter notebooks to BioImage.IO." Maybe the main first description should be also oriented into this.

2.- Contributions: Can you contribute in any other way besides being a developer?

User: feedback as a user. Through communication channels. GitHub chat. You need to sign
Contribute the models to the website (main)
- Contribute weights to existing models.
Contribut to the development of the zoo through GitHub.
Contribute to the runners.
Consumer software.
Bioengine apps.

3.- Model contribution requirements

The models that are upload should run at least in one community partners
In the begining we need to support people's contributions.
How does the developer decide the consumer software? Decision tree with some easy features (TF or PyTorch for example)
The model HAS TO BE in the 0.3.0 format.
Central repo where the developer can do a PR. There is a Bioimage Model repo.
How to contact:
- image.sc (join the community partner? tag: bioimagemodelzoo?)
- github issue
- gitter
What are the requirements in terms of compatibility with community partners to contribute a dataset, a notebook, a model, or a workflow?
- goto the central repo
What if a model cannot be loaded in any of the software? and if I provide the notebook? e.g. from https://github.com/bioimage-io/bioimage.io/blob/master/docs/resource-description-file.md#describing-ai-models, it could be assumed that almost anything can be loaded
- This would be a special contribution. You will always need to provide a model example with the 0.3.0 format.
- Models have attachments? The notebook could be in the attachment.
If I want to contribute with a model, should I create the spec and then ask a partner to include the model in its github with a PR? --> Central repo
Could we provide the link to the repo of each community partner to make the RDF and manifests more accessible?
This info is now at https://github.com/bioimage-io/bioimage.io/blob/master/docs/contribute.md
Description of the RDF and the manifest. Sometimes they seem to be the same and other times, not.

3.- Resource Description File.

The main and basic file to write is the RDF --> link to the description and examples/templates (these can be some datasets or notebooks already available).
Sometimes in the text, it is used "model zoo specification" and others, Resource Description File.
RDF version is 0.2.0. Most probably the entire document should be reviewed: https://github.com/bioimage-io/bioimage.io/blob/master/docs/resource-description-file.md
The name of the RDF should be manifest.bioimage.io.yaml?
What is the main structure of this document and how do I start including my notebooks, datasets, and models?
Section "Describing AI models": how do you feel about merging it with the configuration specification? or make it very clear that there are two kinds of descriptions: for interoperable models and for just models.

4.- Configuration specifications:

The example is missing, the link is broken.
Preprocessing and post-processing reference implementations or mathematical methods.
Run modes specification is still empty. If it is free, then the spec checker should avoid this field. Also, if it is free, shouldn't it go in the config?

5.-Join as a community partner

Information at https://github.com/bioimage-io/bioimage.io/blob/master/docs/join-partners.md
Update the example to 0.3.0. https://github.com/bioimage-io/bioimage.io/blob/master/docs/manifest-format.md

6.- Doc for bioengine-apps

Info at https://github.com/bioimage-io/bioimage.io/blob/master/docs/build-bioengine-apps.md

7.- Template https://deploy-preview-89--bioimage.netlify.app/docs/#/

8.- Tutorials

Videos
Embed the model runner in the page
Potential contacts (who may need help for contributing models)

Outcomes:

The documentation website: https://bioimage.io/docs/
Overview: https://bioimage.io/docs/#/
How-to use Fiji: https://bioimage.io/docs/#/user_guide/how-tos_guide
Tutorials on how to contribute models: https://bioimage.io/docs/#/contribute_models/tutorials
Model runner: https://bioimage.io/docs/#/consumer_software/model_runner
How to join as community partner: https://bioimage.io/docs/#/community_partners/README

oeway commented 3 years ago

March 5th, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Maksim Novikov	EMBL	@m-novikov
Anna Kreshuk	EMBL	@akreshuk
Fynn Beuttenmüller	EMBL	@fynnbe
Daniel Sage	EPFL	@dasv74
Jan Philipp Albrecht	MDC	@jpalbrecht
Florian Jug	CSBD + HT	@fjug

Agenda

Solving the failing tests
- make git_repo optional
- make cite as optional
Zenodo integration to simplify model upload
- Automatic yaml config generation
- Model upload
- tags: bioimage-io for model discovery
Standardisation of the title of the pre-trained mode: some of them include U-Net, or not, some of them highlight the application "Cell segmentation". The front page seems a little weird because the name of models are not unified.
- [TODO] Add recommendation about titles (mind set: what it does NOT what it is (e.g. U-Net))
- [TODO] Update our core model titles
- [TODO] Add recommendation for the covers with image before and after; If it is not image-to-image, the developer should find a way to illustrate whan the model does.
Smaller issues around the website:
- Need a contact button and some e-mail to reach out to stakeholders. Can we have an e-mail at the bioimage.io domain?
  - Add an contact email to the about tab
  - Add issue page to the about tab
- About button should be updated to contain About from docs
  - Update text in the About dialog
- Contribute button is right now showing an old github post instead of actual guidelines
  - redirect user to the contributon in the docs
- The bolding of the model or application word is too subtle
  - update the color

oeway commented 3 years ago

March 12th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Maksim Novikov	EMBL	@m-novikov
Anna Kreshuk	EMBL	@akreshuk
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Florian Jug	CSBD + HT	@fjug
Constantin Pape	EMBL	@constantinpape

Project: https://github.com/orgs/bioimage-io/projects/1

Hackathon: https://github.com/orgs/bioimage-io/projects/2

Agenda

Updates on the website and documentation
Subscription form
Setup the contact form, who want to subscribe, add your name below:
- mamunozb@ing.uc3m.es
- esgomezm@pa.uc3m.es
- florian.jug@fht.org <-- NEW! :)
- janphilipp.albrecht@mdc-berlin.de
- fynn.beuttenmueller@embl.de
- anna.kreshuk@embl.de
Name of the specification
- model spec
Definition of run mode? Multiple run mode?
- Fix spec validator for run_mode
Setup a gather town?
- Free towns can host up to 20 ppl simultaneously.
- Prices can be rather low when you contact them.
- If you have questions, ask Florian (he already wasted a lot of his time to build his gather.town lab... ;)
- https://www.mapeditor.org/
Support exporting zip file from the website (@oeway)

oeway commented 3 years ago

March 19th, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Estibaliz Gómez	UC3M	@esgomezm
Maksim Novikov	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Constantin Pape	EMBL	@constantinpape

Agenda

model packager
- always included into the package:
  - source (single file included as default, additional source files via attachments:files)
  - test_inputs, test_outputs
  - cover
- where to host data files?
web segmentation application demo: https://github.com/Steffen-Wolf/webmutex/tree/master/mutexweb_demo
model card reporting format: https://arxiv.org/pdf/1810.03993.pdf
- Add it as a template for the README file linked to documentation field
Standardization of the names
- name of the model zoo:
  - BioImage Model Zoo
  - bioimage model zoo
  - bioimage-model-zoo
  - NOT: Bioimage Model Zoo
  - NOT: Bioimage model zoo
- name of the domain:
  - bioimage.io
  - bioimage-io
  - NOT: Bioimage
  - NOT: Bioimage.io
  - NOT: Bioimage.io
- A repo name:
  - recommended pattern: XXXX-bioimage-io or bioimage-io-XXXXX
- full name of the spec: BioImage.IO Model Description File Specification
- model.yaml, Acronym: MDF
- BIOM (Bioimage.IO Model)
- repo name of the spec: https://github.com/bioimage-io/spec-bioimage-io
- Package name:
  - bimgio
  - bioimage
  - bioimageio (<< winner)
  - bioimagemodelzoo
  - java: io.bioimage.
- Bioimage, can include biomedical image related model that ONLY for research purpose

oeway commented 3 years ago

March 26th, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CSBD + MPI-CBG	@tomburke-rse
Wei Ouyang	KTH	@oeway
Anna Kreshuk	EMBL	@akreshuk
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz-Barrutia	UC3M	@arratemunoz
Maksim Novikov	EMBL	@m-novikov
Emil Melnikov	EMBL	@emilmelnikov

Agenda

updated model packager
- always included into the package:
  - source (single file included as default, additional source files via attachments:files)
  - test_inputs, test_outputs
  - covers
define consumer spec in the manifest.bioimage.io.yaml
- supported weights formats, version?
- supported pre- and post-processing
- supported spec format version, a list of version numbers
FAQ:
- why not making a weights format convertor?
hosting dataset in zarr format for model prediction

oeway commented 3 years ago

Apirl 9th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Estibaliz Gómez	UC3M	@esgomezm
Maksim Novikov	EMBL	@m-novikov
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Florian Jug	CSBD + HT	@fjug

Agenda

Zenodo integration
- no support zarr: https://github.com/zenodo/zenodo/issues/1599
- alternate: https://www.datalad.org/
- one model per deposition, link to yaml file
- upload first to zenodo, then send PR to add doi to community partner manifest file
- use keywords to discovery new models: bioimageio
Paper with code: https://paperswithcode.com/
Open Problems: https://openproblems.bio/
Model packager file name definition
- the packager adds it
- additional file 'package_info.json' containing pointer to original json file...
- use the file name from url by default, otherwise use weigths without extension.
de.NBI Kubernetes Machines
Java 16 for native access (as if any of you would care...;)
Axes for the models with outputs that are not images (e.g. classification). Related GitHub issue: https://github.com/bioimage-io/spec-bioimage-io/issues/12 (sample vs batch is similar to H/W vs y/x )
idea: add unit for each dimension in input/output tensor description

oeway commented 3 years ago

Apirl 16th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Maksim Novikov	EMBL	@m-novikov
Anna Kreshuk	EMBL	@akreshuk
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Florian Jug	CSBD + HT	@fjug
Tom Burke	CSBD + MPI-CBG	@tomburke-rse

Project: https://github.com/orgs/bioimage-io/projects/1

Agenda

EMBL-EBI course on image analysis for facilities in July (Advanced Users)
- Microscopy data analysis: Machine learning and the BioImage Archive
- https://www.embl.de/training/events/2021/BIA21-01/programme/index.html
- Ask ZeroCost to add a part about exporting model for the bioimage model zoo
- TODO: talk to Zero ask for support for exporting model.yaml
- Make model package and use them in different consumer tools.
- How to setup the infrastructure
discussion about the new axes: https://github.com/bioimage-io/spec-bioimage-io/pull/69
- metadata
  - option 1:
    - axes: "btxyz"
  - option 2:
    - axes: ['batch', 'time', 'x', 'y', 'z']
implement the packager in Python for generating package and validate (May?)
- ilastik team will implement it until May 7
testing deNBI cloud
- setup a K8S cluster and install BinderHub
- ImJoy AI Engine with S3 storage? for serving zarr/NGFF datasets, collecting annotation, support interactive annotation
- how to host dataset, start with xarray?

oeway commented 3 years ago

Apirl 23rd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Anna Kreshuk	EMBL	@akreshuk
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Tom Burke	CBG+CSBD	@tomburke-rse

Agenda

Settle down the axes
- Make a PR for adding i as an axes
- make anotehr PR for change the axes as a list of dictionaries
- postpone the discussion about allowing duplicating channel
- postpone the discussion about adding test_tensor

oeway commented 3 years ago

Apirl 30th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Estibaliz Gómez	UC3M	@esgomezm
Constantin Pape	EMBL	@constantinpape
Tom Burke	CBG+CSBD	@tomburke-rse
Joran Deschamps	MPI-CBG	@jdeschamps
Anna Kreshuk	EMBL	@akreshuk
Florian Jug	CSBD + HT	@fjug	-->

Agenda

Spec format? validator? (example generated md doc: https://github.com/bioimage-io/python-bioimage-io/blob/6483886f0770c0157a77880661824314caab0ba0/generated/bioimageio_model_spec.md)
All: use cases for the preprint
Strategy for models that do not meant to be cross-compatible. E.g. CellFinder, DeeplabCut?
- the strategy is to just require the RDF for these models
- connected to this: we want the specs to live in the same location: https://github.com/bioimage-io/spec-bioimage-io
- for RDF(s) (are these different for notebooks, dataset, model?) these will be jsonschema
- for the full model spec, it will be defined in marshmallow (see point1), hence the python code will also be migrated to the spec repository
- the spec files will then be used to generated the documentation for the website (https://bioimage.io/docs/#/)
Implementing the test run button on the webstie
- deepimagej -- blocked on the headless mode
- fiji -- Tom is working on it
- ilastik -- work on the tiktorch model runner
[meeting at 4:30] Zero: export model from notebooks prior to preprint submission...
- One use-case to work on (multi-consumer use-case):
  - 3D U-Net, tained using Zero, exported in Model Zoo format (via Esti's magic using TF zip) -- DONE!
    - Dataset ideas:
      - arabidopsis dataset https://osf.io/w38uf/
      - CREMI dataset: https://cremi.org/data/
      - One of the EmbedSeg ones?
  - Consumed in ImageJ via DeepImageJ and output used downstream for something.
  - Consumed in ilastik, e.g. for multi-cut or so.
  - Consumed by BioEngine (or Imjoy?) to test on some user data on BioImage.IO.
- Use-case two: StartDist/ N2V
  - Train StarDist in Zero
  - Persuade Uwe to become ModelZoo compatible
  - TODO Anna: talk to Uwe please!
  - Important selling point: BioEngine makes it work online too!
  - Retraining of stardist in zerocost from a stardist model in the model zoo
- Use-case three: Image Classification
  - Producer: last Kaggle challenge
  - Consumers: DeepImageJ and ImJoy
Florian: Trusted-tier vs. community contribution vs. unlisted
- multi-consumer vs single-consumer
- full spec vs minimal spec
- --> moved to next Friday (I hope I can be there, but likely I will...)

oeway commented 3 years ago

May 21st, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Constantin Pape	EMBL	@constantinpape
Tom Burke	CBG+CSBD	@tomburke-rse
Anna Kreshuk	EMBL	@akreshuk
Estibaliz Gómez	UC3M	@esgomezm
Emil Melnikov	EMBL	@emilmelnikov
Dominik Kutra	EMBL	@k-dominik

Use cases (copied from last meeting)

One use-case to work on (multi-consumer use-case):
- 3D U-Net, tained using Zero, exported in Model Zoo format (via Esti's magic using TF zip) -- DONE!
- Link to the notebook: https://github.com/HenriquesLab/ZeroCostDL4Mic/blob/master/Colab_notebooks/Beta%20notebooks/U-Net_3D_ZeroCostDL4Mic_BioImageModelZoo_export.ipynb
  - Dataset ideas:
    - arabidopsis dataset https://osf.io/w38uf/
    - CREMI dataset: https://cremi.org/data/
    - One of the EmbedSeg ones?
- Consumed in ImageJ via DeepImageJ and output used downstream for something. (@Esti train the model and try it in deepimagej)
- Consumed in ilastik, e.g. for multi-cut or so.
- Consumed by BioEngine (or Imjoy?) to test on some user data on BioImage.IO.
Use-case two: StartDist/ N2V
- Train StarDist in Zero (@Tom 1-2 weeks)
- Important selling point: BioEngine makes it work online too! (@Wei )
- Retraining of stardist in zerocost from a stardist model in the model zoo
Use-case three: Image Classification
- Producer: last Kaggle challenge
- Consumers: DeepImageJ and ImJoy (@esti )

Agenda

machine learning group @ elixir https://elixir-europe.org/focus-groups/machine-learning
- DOME https://dome-ml.org/
Wei:
- deNBI cloud setup suspended due to onging mantainance
- setting up k8S/storage server on embassy cloud
- zarr/virtual stack support for ImageJ.JS
Model upload, what's the current status?

TODOs for Wei:

Implement the zenodo upload button
Convert the 3D Unet zero notebook into an bioengien app

oeway commented 3 years ago

BioImage.IO Meeting Minutes

May 21st, 2021

:::info Time: 3:00PM CET (your timezone) Where: Zoom Channel :::

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Constantin Pape	EMBL	@constantinpape
Anna Kreshuk	EMBL	@akreshuk
Estibaliz Gómez	UC3M	@esgomezm
Florian Jug	CSBD + HT	@fjug

Project: https://github.com/orgs/bioimage-io/projects/1

Agenda

Wei: Upload feature demo: https://github.com/bioimage-io/bioimage.io/pull/104
- Anyone can upload models tagged with bioimageio through our bioimageio client or manually to zenodo, will be rendered as community items by default
- To add to be listed as a trusted items, uers needs to send a PR
Florian: Trusted-tier vs. community contribution vs. unlisted
- multi-consumer vs single-consumer
- full spec vs minimal spec

oeway commented 3 years ago

June 11st, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Anna Kreshuk	EMBL	@akreshuk
Dominik Kutra	EMBL	@k-dominik
Florian Jug	CSBD + HT	@fjug
Uwe Schmidt	None	@uschmidt83
Martin Weigert	EPFL	@maweigert
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz

Agenda

Stardist: Use case discussion
- https://github.com/stardist/stardist/pull/149
- Suggestion: Make it clear what is the truth of the spec, document it well in the README
Generic and consistent entry point:model.yaml ==> rdf.bioimage.io.yaml in the zip package
- suggestion: rdf.yaml
- (this will be part of 0.4)
- Debugging with folder
Only allow files, not subfolders (0.4)
RDF type collection allows linking a list of RDFs (deprecating manifest file format)
- unify format version of RDF and model spec (issue)

oeway commented 3 years ago

June 18th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Anna Kreshuk	EMBL	@akreshuk
Dominik Kutra	EMBL	@k-dominik
Florian Jug	CSBD + HT	@fjug
Estibaliz Gómez	UC3M	@esgomezm
Jan Philipp Albrecht	MDC	@jpalbrecht
Tom Burke	CBG+CSBD	@tomburke-rse

Agenda

The course on July 16th, about BioModel zoo for the users: https://www.ebi.ac.uk/training/events/microscopy-data-analysis/
- The users will try out at least one use case of the ones above
- The most advanced ones can go as far as fine-tuning in ZeroCost again
- Maybe show the 3d use case and then have them discover the 2d by themselves
Use zenodo as the default way of hosting official models?
- Zenodo api --> github CI (filtering transforming RDFs) --> JSON --> Render on site --> packager
bioimage-io community, approval, github bot 👍
Napari grant
- enable developers to convert an existing image analysis tool to a napari plugin;
- ~~sustain and continue development of an existing napari plugin;~~
- or build foundational infrastructure for other plugin developers.
Discussions about website design
The packager should just be called "Download" on the website, because that's more intuitive for users. It should also have a download symbol.
Weight formats need to be more understandable. For this, we need to add a json that maps the names of weight formats to a more understable description. This should contain the information about which software supports the model. UI idea: checkboxes for the different consumers and the correct weight formats are then selected. THere is a pro-mode that let's you select explicit weight formats. To make this happen we need the list of supported weight formats for each consumers, including their priorities, e.g. ilastik: pytorch-state-dict, torchscript, onnx. Where should this information be stored? In the consumers manifests?
Right now, the user needs to select the weight format. Should we add an option for "all"?

Discussion about default model format

Longer discussion if we should support both the "old" way of hosting models via github and the manifest.yaml or only support the "new" way of hosting via zenodo. We agreed to only going for zenodo right now to keep things simple. Eventually it should be easy to setup some CI that moves github hosted models to zenodo / updates them. Add the icons for consumers via the "Link" functionality. Eventually we want to auto-detect the supported links (supported consumers). For this, consumers should eventually provide some function that checks the validity for the consumer, in the plugin app.

oeway commented 3 years ago

June 25th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Anna Kreshuk	EMBL	@akreshuk
Dominik Kutra	EMBL	@k-dominik
Jan Philipp Albrecht	MDC	@jpalbrecht
Tom Burke	CBG+CSBD	@tomburke-rse

Agenda

The workflow tab, merge notebooks for workflows? since notebooks are just worflows that genreates a model.
Sorting strategy
next minor version (0.4.0): update tensor description, update weight formats
- axes?
- Github discussions
- Model RDF vs generic RDF https://github.com/bioimage-io/spec-bioimage-io/issues/140
- Consumer on model card
- Models on zenodo; who to give multiple maintainers access?
- Model downloader for ilastk
- Customize https://github.com/imjoy-team/bioimage-io-models/blob/master/src/bioimageio-packager.imjoy.html

For generic RDF spec
- Fix authors
- Add links
  Workflow vs. Notebooks
merging "Notebooks" and "Workflows" is a good ideas
"Workflows" are then a good place to document the training, e.g. notebook, python script, etc.
Also for other workflows like complicated post-processing etc.
Could think about specializing the rdf for workflows a bit more, e.g. enforcing tags

oeway commented 3 years ago

July 2nd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz
Dominik Kutra	EMBL	@k-dominik
Tom Burke	CBG+CSBD	@tomburke-rse

Agenda

Updates on the website: edit button, authors and cite field, consumer bioengine app example.
Tew modes of test run button: browser computing and cloud computing. Web ML loader api
Feedback from the ilastik course. Considering we have to deliver another training on the 16th, I would propose the following plan of action, to be discussed: 1) We need to fix the model spec RIGHT NOW. I believe there are very few points left, let’s come to a decision. As an additional point, the general rdf and the model rdf should be identical in the overlapping fields. 2) All models on the website have to follow the fixed spec 3) Uploaded models are checked against the spec (Fynn suggested it could be done through ImJoy) 4) All end tools support the fixed spec. Ideally, there is CI which checks the combo of packager+tool at least for loading the model. We will definitely do it for ilastik, but I have no idea how hard it is for others.
Debug/testing/staging environment: Currently everything is deployed to production right away. It would be nice to have a more gradual approach to this - preprod.bioimage.io?

Notes

Python-based packaging (bioimage.spec): We'll plan on integrating this in a browser-local (pyodide?) app, do the check and packaging all from the same library. Todo: open issue.
General RDF 0.3.0: sync with model
With the switch from gh-based hosting of models to zenodo, manifest.bioimage.io.json will pool from zenodo - for this it is necessary to get direct download links for the packaged versions. One possibility would be to setup CI that checks models on zenodo, and runs the packager - stores produced binaries on github-releases. There might be an issue with tracking of download numbers and the likes from zenodo.
Aim for release of 0.3.2 version of the spec by the end of next week.
for all our repos: rename branch 'master' to 'main'
we encourage to consider the persistency of links used in the context of bioimage.io pointing to 'main' branches can lead to unintended dynamic changes, tags or hashes are alternatives, but bring their own disatvantages. We will not enforce any particular github link style across bioimage.io.

oeway commented 2 years ago

July 9th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz
Cristina de la Torre	UC3M	@crtorreg
Tom Burke	CBG+CSBD	@tomburke-rse
Jan Philipp Albrecht	MDC	@jpalbrecht
Emil Melnikov	EMBL	@emilmelnikov
Dominik Kutra	EMBL	@k-dominik
Anna Kreshuk	EMBL	@akreshuk

Agenda

Course next Friday
- Which models do we need, torchscript models for joint application in ilastik deepImageJ?
- Can we release 0.3.2 now? Is something still missing? Unifying fields (authors, attachments) in model and general RDF, or in next release?
- Can we set up CI for the models on zenodo till then? (Currently this is the biggest point missing to make everything more stable.)
- What's the issues left for the website?
  - Add a button for contact us, fix the contact us
  - Ask a question image.sc?
General discussions (would be good to cover the basics to have more informed discussions on gh after that)
- Attachments: https://github.com/bioimage-io/spec-bioimage-io/issues/148
- Dataset RDF: https://github.com/bioimage-io/spec-bioimage-io/issues/153
- spec configuration: https://github.com/bioimage-io/spec-bioimage-io/issues/156

Comments

Information to render the models in the web page: https://github.com/bioimage-io/collection-bioimage-io/blob/gh-pages/rdf.yaml
The old manifest.bioimage.io.json (https://github.com/bioimage-io/bioimage-io-models/blob/gh-pages/manifest.bioimage.io.json) will not be deleted but it will not be updated.

Discussions

for the general discussions we agreed with the less restrictive checks
for the next release we will work on general RDF 0.3.0, where we will discuss how to make things more restrictive and discuss more general what needs to be defined as RDFs and what is implementation detail that should not be part of our specs

oeway commented 2 years ago

July 16th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz
Cristina de la Torre	UC3M	@crtorreg
Emil Melnikov	EMBL	@emilmelnikov
Anna Kreshuk	EMBL	@akreshuk
Daniel Sage	EPFL	@dasv74
Florian Jug	HT	@fjug

Project: https://github.com/orgs/bioimage-io/projects/3

Agenda

Feedback from the course
- Should we support analysis tools/scripts in general
  - Yes! IN THE FUTURE! ;)
- Allow downloading datasets?
  - Spec for define how to download, combined with bioengine apps
  - https://github.com/bioimage-io/spec-bioimage-io/issues/153
  - https://github.com/bioimage-io/bioimage.io/issues/147
- How many models do we have? Generate a statistics page would be great
- A python API for list and run the models?
  - -> python-bioimage-io?
  - bioimageio.runner?
- Show version history of every model
- Show user statistics of every model
- Zenodo upload API: https://developers.zenodo.org/#introduction
  - bioimageio.uploader?
  - bioimageio.downloader?
  - part of bioimageio.spec?
Plans
- Keep the momentum!
  - 1-day hackathon
- Make a list for the release
  - Add list here: https://github.com/orgs/bioimage-io/projects/3
CZI napari plugin grant???
- YES!
  - Who about what? (JugLab is also writing some... ;)
  - Lundberg lab is working on a draft
    - 20k (translation/maintainance) proposal or 50k (infrastrcuture) proposal? the later(?) -- ok... because I think also the 50k one would work well. -- we can coordindate
    - Who will be the 3 PIs on it?
  - EMBL (Constantin) & MRC (Steffen Wolf) are also working on one, focus is interactive mutex watershed segmentation, would be Translation and use model zoo
  - Wei (I hope?!)
Suggestion from Guillaume Jacquemet(ZeroCost)): For the dataset part, it would be good to be careful what the "by" mean. Is it "entry by" or dataset created by?
FYI: BioImage.IO icon/logo/diagrams:
- the icons and logos files are here.
- diagram on how it works: https://docs.google.com/drawings/d/154letH370zIzjNs36YCpeC19aMiZzTnq74TtDm8GSw8/edit?usp=sharing

FOR THE PREPRINT RELEASE: PLEASE ADD THE REMAINING ITEMS TO HERE: https://github.com/orgs/bioimage-io/projects/3

oeway commented 2 years ago

Friday Hackathon

July 23rd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Fynn Beuttenmüller	EMBL	@fynnbe
Constantin Pape	EMBL	@constantinpape
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Emil Melnikov	EMBL	@emilmelnikov
Dominik Kutra	EMBL	@k-dominik
Anna Kreshuk	EMBL	@akreshuk
Steffen Wolf	MRC LMB	@Steffen-Wolf

Agenda

Use Case 1:
- try 2D/3D U-Net exported from zero cost in ilastik
- bioengine app and/or hybrid desktop: create/reactivate the bioengine app for ilastik and deepimagej
Refactoring python libraries:
- draft structure of python-bioimageio and relation to spec-bioimageio
- use python-bioimage-io in ilastik bio engine app (or current tiktorch)?
CI! pull models from zenodo...
- check spec (with spec-bioimage-io)
- check test input/output (with python-bioimage-io)
- cache results and model overview ("gh-pages"?)
- Run spec-checker for the zenodo
MutexWatershed BioEngine Plugin
- make onnx work github pages
- port it as a webworker plugin in ImJoy (https://imjoy.io/lite)
- Connect with the BioEngine API(accept models from the bioimage.io website)
- Demo: https://imjoy.io/lite?plugin=https://raw.githubusercontent.com/oeway/webmutex/master/runMWS.imjoy.html https://imjoy.io/lite?plugin=https://raw.githubusercontent.com/Steffen-Wolf/webmutex/master/docs/runMWS.imjoy.html
Ref:
- the viewer: https://kitware.github.io/itk-vtk-viewer/app/
- ndarray encoding: https://github.com/imjoy-team/imjoy-rpc
- How to turn a website to a ImJoy plugin: https://github.com/imjoy-team/imjoy-core/blob/master/docs/integration.md#use-your-web-application-inside-imjoy
- Enable dev mode in bioimage.io: https://bioimage.io/#/?dev=1
- Connect to bioEngine: https://github.com/bioimage-io/bioimage.io/blob/main/docs/bioengine_apps/build-bioengine-apps.md
Documentation
- Some documentation for user is missing. We will create a "Welcome tour slides" using ImJoy.
  - The files should saved here: https://github.com/bioimage-io/bioimage.io/tree/main/docs
  - ImJoy Slide: https://github.com/imjoy-team/imjoy-slides --> click on https://slides.imjoy.io/?edit=1 to start creating the slides interactively (it's markdown).
  - To see the slides: https://slides.imjoy.io/?slides=URL where URL points to the markdown.
- Folder with the Figures to copy them in a public google drive: https://drive.google.com/drive/folders/1DHyq2N1ee6Abvu3Ya2TsoRcy08WQDADJ?usp=sharing
- Wei: pop up dialog for the first time they use the model zoo or a model for example.

Discussions:

python library: Fynn, Constantin, Dominik
MWS BioEngine: Constantin, Steffen, Wei
Use case 1: after trying in ilastik Esti, Wei, Constantin
Wrap up meeting at 5pm

Most important points for the preprint:

Not just a collection, but we also have spec checker, and CI to test run
Anna's:
- We need something stable
- Ensure that all model in the zoo run in all the consumers in which they are tagged.
- CI --> really important
- Idea for the preprint: Brief communications (2 pages of correspondence, ~ 1800 words + 2 figures)
- The big claim for the spec is that there's no a random collection, they are checked
- Ensure 3 use cases.
- The release should show solid foundation, to demonstrate its is reliable.

Hackathon next Friday

TODOs:

CIL: cache zenodo; add spec validation; add input/output test: Wei, Dominik, Fynn
Switch to Zenodo production server

oeway commented 2 years ago

August 27th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Estibaliz Gómez	UC3M	@esgomezm
Arrate Muñoz	UC3M	@arratemunoz
Cristina de la Torre	UC3M	@crtorreg
Tom Burke	CBG+CSBD	@tomburke-rse
Anna Kreshuk	EMBL	@akreshuk
Florian Jug	HT	@fjug

Agenda

Preprint plan
- Wei & Esti will start to pull text together and make drafts for figures, then circle around
Tiling discussion
- we should move on with our current spec (since we already have halo and min)
- if necessary we can add more keys to instruct software to do different tiling
Tensorflow in ImageJ
- The best way forward we think is to use pyimagej and leave deep learning in Python
Support production model in the BMZ
Datasets supports in BMZ
- Potential solution for dynamic data serving and conversion, similar to the model packager

oeway commented 2 years ago

September 3rd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Estibaliz Gómez	UC3M	@esgomezm
Dominik Kutra	EMBL	@k-dominik
Cristina de la Torre	UC3M	@crtorreg
Constantin Pape	EMBL	@constantinpape

Agenda

Is the java library up-to-date with the spec?
- Tom: There is a PR missing, will try to work on it next Monday
Collabration with QuPath: Pete tried with early version, pixel size is missing. We will meet in a month.
Preprint & Release plan
- StarDist need the python library to load the model
- Make the preprint more accessible for biologist, idea: tell a story with persona and exampes to describe the life cycle of models
The bioimage-core python library will be released next week: support pytorch/tensorflow/onnx

oeway commented 2 years ago

September 10, 2021

Attendees

Name	affiliation	username (github)
Tom Burke	CBG+CSBD	@tomburke-rse
Estibaliz Gómez	UC3M	@esgomezm
Dominik Kutra	EMBL	@k-dominik
Cristina de la Torre	UC3M	@crtorreg
Constantin Pape	EMBL	@constantinpape
Jan Philipp Albrecht	MDC	@jpalbrecht

Agenda & Discussion

zero-cost & tensorflow
- testing the zero cost model: https://github.com/bioimage-io/spec-bioimage-io/issues/195
- add keras, tf, tf-js example model: https://github.com/bioimage-io/spec-bioimage-io/issues/111#issuecomment-916745219
  - https://github.com/deepimagej/tensorflow-js-models
imagej release 0.3.2: done.
plan for python libraries: release next week

oeway commented 2 years ago

September 17th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Anna Kreshuk	EMBL	@akreshuk
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Arrate Muñoz	UC3M	@arratemunoz

Agenda & Discussion.

Python libraries update
- copy/mark_reference processing
- @wei TODO: Implement pypi release CI for core library
Next spec version:
- Axes
- Fix two way to define output shape
- No duplicate tensor names across inputs/outputs: https://github.com/bioimage-io/spec-bioimage-io/pull/236
Raw nodes vs nodes: bioiamge.io core
Model serving, KubeFlow, Triton Inference server & KFServing:
- predictor: A model
- transformer: a docker container with input and output args
- @wei Try to deploy DSB model: https://bioimage.io/#/?tags=dsb&id=10.5072%2Fzenodo.881940

oeway commented 2 years ago

September 24th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm

Agenda & Discussion.

AI4Life grant application submitted
Added the HPA models and used it as test case (thanks to Constantin)
Model spec verification before uploading
Model converter for serving
Demo: Model serving + JupyterLite: CellPose and HPA
Releasing python stuff
Test models for keras / tf

oeway commented 2 years ago

October 1st, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Constantin Pape	EMBL	@constantinpape
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg

Agenda & Discussion.

Esti: feedback from ZIDAS
- The users were mostly biologists. The teachers were computer scientist. There doubts were:
- Where is the data to train the model?
- All the models have a way to report a github issue, where the issue goes? How can I track my issue? Maybe is a good idea to forward them to the forum. Wei argue that there is already a small forum for each model. You can follow the threat of your issue there. Wei propose that the model developers that upload the model give a mechanism to communicate the doubts to them. Conclusion: configurable issue generation combining it with the image.sc forum (with the tag of the model).
- Most of the doubts where about the appearance of the interface
- Too many download options, esp. "Fiji" vs. "DeepImageJ" is confusing in the Model packager, users often chose Fiji when trying to download deepImageJ models.
  - Suggestion: replace "Fiji" by the plugin name (remove it here: https://github.com/bioimage-io/spec-bioimage-io/blob/gh-pages/weight_formats_spec.json#L195)
  - We need to communicate with Florian and Tom what is the current status with CSBDeep Fiji plugin
  - Namespacing for consumers that have different plugins, e.g. startdist-fiji and stardist-python, because they are not (fully) compatible
- In the BMZ, if the model doesn't work, how to finetune models?
- Make it easier to find models
  - maybe make search fuzzy
- They do not like to click and download, drag and drop. Maybe installation through the id.
  - This should be the recommended way of getting models!
  - Add a button to copy the id from the model and this should then be copy-pasted in the consumer (where currently the path goes)
Meeting with Pete from QuPath team
- Add petebankhead to the java core libary
  - Yes to adding Pete, he can help in developing, maintaining the library
- Move the java core library to the bioimageio org? (need to discuss with Florian and Tom)
- would also be interested in using the python library for creating and/or packaging models
- CP will add some documentation on usage and also share with Martin Weigert
DVC and CML for model and dataset versioning: https://dvc.org/
4x GPUs
BioEngine / ImJoy App Engine development
- [Prototype] Serverless app deployment (imjoy plugins running in the server side)
- [WIP] virtual file system support (based on fsspec)
- [WIP] Experimental inference/training env support, the following are required
  - conda environment.yaml (required to be submitted to an official ML conda environment repo)
  - a model entrypoint file with the following function:
    - initialize load the model
    - execute run the prediction or training for a few epochs
    - finalize (optional) to unload the model
  - bioengine application files (ImJoy plugins, e.g. js or pyodide) for user facing applications, e.g. inference app, training app
[TODO] making sure that models work
- Validation when updating models (Wei will implement)
- We need to get the CI testing models up and running again, CP will check

oeway commented 2 years ago

October 8st, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Anna Kreshuk	EMBL	@akreshuk

Agenda & Discussion.

Validation and CI
- Spec validation when uploading models
- CI for validating models:
  - Reactivate CI that validates and tests models, updating the batches
  - Check that models exist (sometimes they just vanish?!?) and more website tests, e.g. checking tags
BioEngine / ImJoy App Engine development
- [WIP] Experimental inference/training env support, the following are required
  - conda environment.yaml (required to be submitted to an official ML conda environment repo)
  - a model entrypoint file with the following function:
    - initialize load the model
    - execute run the prediction or training for a few epochs
    - finalize (optional) to unload the model
  - bioengine application files (ImJoy plugins, e.g. js or pyodide) for user facing applications, e.g. inference app, training app
Deploy flask / fastapi web apps
packager issue (https://github.com/bioimage-io/bioimage.io/issues/171)
copyable doi for bioimageio cards
Updates on python library
- doi support is coming
- prediction with tf1 is working, keras still needs to be tested, @Esti did you have a chance to look into this?
- usage examples WIP
  - need to fix https://bioimage.io/#/?tags=DSB-Nuclei-BoundaryModelNew&id=10.5072%2Fzenodo.881940

oeway commented 2 years ago

October 15st, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Lucia Moya	UC3M	@XXX
Anna Kreshuk	EMBL	@akreshuk
Constantin Pape	EMBL	@constantinpape
Florian Jug	HT	@fjug

Agenda & Discussion.

Updates to the website, new features:
- Run validator before uploading
- Copiable id
Bioimageio core example notebook: https://github.com/bioimage-io/core-bioimage-io-python/blob/main/example/bioimageio-core-usage.ipynb
Interoperability, running zero-cost models in python
CI for testing models
Hackathons et al.?
- idea for filling up the bioimage model zoo
- Let's do it!
- The week of Nov. 29th!
Warning: Currently we are in sandbox mode!
Question: Preprint?

oeway commented 2 years ago

October 22nd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Jan Philipp Albrecht	MDC	@jpalbrecht
Arrate Muñoz	UC3M	@arratemunoz
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Lucia Moya	UC3M	@XXX
Anna Kreshuk	EMBL	@akreshuk
Constantin Pape	EMBL	@constantinpape
Jan Philipp Albrecht	MDC	@jpalbrecht

Agenda & Discussion.

When can we do a feature freeze?
- CI is being set up; @esti, how is that going on your side; example model for keras/tf
- DeepImageJ CI is working
- Model Validation on upload not working?
hackathon planning
- Zero-cost support import and export
- ilastik
Updates for the deNBI cloud: the triton server is running (done by Craig)

oeway commented 2 years ago

October 22nd, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Constantin Pape	EMBL	@constantinpape
Jan Philipp Albrecht	MDC	@jpalbrecht

Agenda & Discussion.

deepImageJ: https://github.com/bioimage-io/spec-bioimage-io/pull/267
- runmode: deepimagej to skip the model test
- example model (DeepSTORM): https://bioimage.io/#/?tags=deepstorm&id=10.5072%2Fzenodo.906808
- To avoid using non-defined postprocessings, we try the combination model + aplication. Aplication would be the imagej macro calling the model and performing the complex postprocessing.
- This approach allows defining models in a simpler manner: the model does a specific task (segmentation, denoising ...) and you integrate this specific task-performance in the pipeline that gives you the desired result (for example, watershed for instance segmentation)
- Sometimes users mess tasks and models so they just try for example CARE for segmentation and a UNet defined for training for denoising. Keeping it simpler, may contribute to show a clearer idea.
CI for model validation
- added in https://github.com/bioimage-io/collection-bioimage-io/pull/7
- categorize errors into: static and dymanic (per weight format)
- make it json output
- single file for model
(Esti: I left but I let it here) Is there any plan for the hackathon? How many people? timing? This would be important to organize the rooms and all that in Madrid.

oeway commented 2 years ago

Nov. 5th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Estibaliz Gómez	UC3M	@esgomezm
Anna Kreshuk	EMBL	@akreshuk
Constantin Pape	EMBL	@constantinpape

Agenda & Discussion.

Hackathon planning: https://docs.google.com/spreadsheets/d/1TXUx8QAldBaQLrVkBUgjMKnPxQv3gLSWExoAMcFTRig/edit
- Preprint release project: https://github.com/orgs/bioimage-io/projects/3
  - Add todo items to the project (use Hackathon 29th Nov)
  - Until next meeting, add the missing todo items
  - Group/clean up the list during the next meeting
  - Goal: release the website, documentation, models, CI, spec, core library, preprint
  - Coordination with zerocost, stardist, qupath
    - TODO: notify qupath about the hackathon
    - TODO: notify zerocost about the hackathon (add notebook cells for importing and exporting bioimage models)
StarDist integration update
- https://github.com/stardist/stardist/pull/171
- question for esti: how to store parameters for postprocessing such that it can be used for deepimagej
vscode web, bioengine and model serving

oeway commented 2 years ago

Nov. 12th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Tom Burke	CBG+CSBD	@tomburke-rse
Constantin Pape	EMBL	@constantinpape
Dominik Kutra	EMBL	@k-dominik
Fynn Beuttenmüller	EMBL	@fynnbe
Jan Philipp Albrecht	MDC	@jpalbrecht
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Carlos García	Pasteur	@carlosuc3m
Ricardo Henriques	IGC	@HenriquesLab

Agenda & Discussion.

Hackathon planning:
- https://docs.google.com/spreadsheets/d/1TXUx8QAldBaQLrVkBUgjMKnPxQv3gLSWExoAMcFTRig/edit#gid=0
- GH project: https://github.com/orgs/bioimage-io/projects/3
- zerocost:
  - Support a subset of the notebook
  - Link back from the zerocost github wiki
  - Help with the document
Processing tests across consumers https://github.com/bioimage-io/core-bioimage-io-python/issues/145
- Generate tests inputs and outputs for pre/post-processing
- Add documentation
Adding bioimage.io to the triton server
- mixed pytorch/tf environment example: https://github.com/bioimage-io/collection-bioimage-io/blob/main/dev/environment-mix.yaml
- we should probably pin some dependencies in this environment to ensure the triton server and the CI test env are compatible
Stardist export is working https://github.com/stardist/stardist/pull/171
pre-postprocessing docs are generated now: https://github.com/bioimage-io/spec-bioimage-io/tree/gh-pages

oeway commented 2 years ago

Nov. 19th, 2021

Attendees

Name	affiliation	username (github)
Wei Ouyang	KTH	@oeway
Constantin Pape	EMBL	@constantinpape
Lucia Moya	UC3M	@XXX
Dominik Kutra	EMBL	@k-dominik
Jan Philipp Albrecht	MDC	@jpalbrecht
Estibaliz Gómez	UC3M	@esgomezm
Cristina de la Torre	UC3M	@crtorreg
Carlos García	Pasteur	@carlosuc3m
Arrate Muñoz	UC3M	@arratemunoz
Carlos García	UC3M	@carlosuc3m

Agenda & Discussion.

Hackathon planning: https://docs.google.com/spreadsheets/d/1TXUx8QAldBaQLrVkBUgjMKnPxQv3gLSWExoAMcFTRig/edit#gid=0
- Please add meeting schedules below the form
Updates from zenodo: CORS issue (fixed) and range request support
- zarr sharding for consolidating multiple chunks into one file: https://github.com/zarr-developers/zarr-python/issues/877#issuecomment-973772122
Training support for the BioEngine
StarDist integration is on the way, compat. with deepimagej
- Generate deepimagej config in bioimageio.core
- Martin & Uwe will join hackathon Thursday afternoon
  - Example deepimagej model: keras works! tf not yet

bioimage-io / bioimage.io