tinevez
commented
1 year ago Hello @jiyuuchc
We wrote a TrackMate detector plugin. This is a deep learning based cell detector, originally developed for a local hematology research project, but we are considering releasing it to the public. While Trackmate already includes several cell-detector options, some users might still find our alternative useful due to its speed.
Awesome! We really built TrackMate so that it can be extended by others, and this issue makes my day! I can't wait to study the technology you used to communicate with Python.
So before we upload it to image.net, we would like to double check with the upstream to see if there is any recommended "best-practice" for doing this, e.g., just off the top of my head:
- Package name: we currently use ij.plugin.trackmate.lacss, which in theory could lead to collision. Should we change that?
I would recommend using the same root that of the TrackMate packages, though:
fiji.plugin.trackmate.detection.lacss
so that it fits nicely under the same package hierarchy common to all detectors.
- Artifact name. Any conventions?
We recommend TrackMate-Something, with a capital 'S'. See for instance the artifact names for the other extensions:
https://github.com/orgs/trackmate-sc/repositories
- Is it ok to hijack trackmate's mailing list tag, e.g. https://forum.image.sc/tag/trackmate, for the plugin's POM metadata?
Totally ok.
- Anything else?
We are working on a v8 version, that will support 3D segmentation. This will cause major changes in the API that might affect you. Check the draft PR for a succint description of what will happen: https://github.com/trackmate-sc/TrackMate/pull/263
But in any case we will be there to help with porting your code to the future new v8, to be out probably early next year or for Christmas.
jiyuuchc
We wrote a TrackMate detector plugin. This is a deep learning based cell detector, originally developed for a local hematology research project, but we are considering releasing it to the public.
While Trackmate already includes several cell-detector options, some users might still find our alternative useful due to its speed.
So before we upload it to image.net, we would like to double check with the upstream to see if there is any recommended "best-practice" for doing this, e.g., just off the top of my head: