OME Zarr hackathon 2022

OME-Zarr hackathon 2022

John Bogovic's plan for the Bioimage hackathon 31.01.2022 - 06.02.2022

Transform specification

Views and spaces

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
• what are good defaults / assumptions ?
• Should implementations be able to handle implicit concatenation
  • if there are two transforms specified: T: A -> B and S: B -> C, is ST: A -> C understood to exist implicitly?

New transform types

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.

Implementation

• 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