-
Notifications
You must be signed in to change notification settings - Fork 8
OME Zarr hackathon 2022
John Bogovic edited this page Feb 1, 2022
·
5 revisions
John Bogovic's plan for the Bioimage hackathon 31.01.2022 - 06.02.2022
Work on the "spaces" and specify transforms between spaces. Start of work is here.
Main idea:
- A "space" is a set of "axes"
- A "coordinateTransform" is a function from one space to another.
Identify some relevant use cases to show why this is a useful idea, and as a guide to specify it. Here are a few:
- "raw" / "array" space (discrete)
- raw to physical
- crops of a larger volume
- lower-res to higher-res in a multiscale pyramid
Some open questions
- What is the right place to store transforms and spacing?
- transforms may need to know about many spaces corresponding to different datasets
- first thought is to keep them together in one central place'
- application specific - Saalfeld
- what are good defaults / assumptions ?
- Should implementations be able to handle implicit concatenation
- if there are two transforms specified:
T: A -> B
andS: B -> C
, isST: A -> C
understood to exist implicitly?
- if there are two transforms specified:
Many transform types I initially proposed here are not in v0.4.
- Revisit which of these should be considered in future versions.
- Consider adding new types not in that list?
- This is more a question for the broader community
- Consider use cases, stories, and examples for proposed transform types.
- Start implementing in imglib2 / bdv
- look for volunteers for a python implementation
- Look again at sharing implementation between MoBIE + N5-viewer
- Talk with Tobias + Tischi + Constantin about what repo might be an appropriate location for this