search

jpdefrutos / DDMR

3D image registration training framework using adaptive loss weighting and synthetic data generation
MIT License
7 stars 2 forks source link

Make simple CLI for deployment #8

Closed andreped closed 11 months ago

andreped commented 1 year ago

It would be great for researchers interested in trying our final models, if it was possible to run the model easily on their own data, from original NIfTI to final predictions stored as NIfTI.

You could take inspiration from what was done in the livermask tool: https://github.com/andreped/livermask#usage

Instead of going through PyPI or enforcing pip to install the tool (as done here), we could make executables, which would be more user-friendly. For instance as done for Raidionics (see here).

The only thing I need is a working main.py which takes a fixed and moving image and produces a registered image. You could also add support to register segmentations, if you'd like.

Then we can setup a .spec-file which should yield an executable (see here). The final installer would need to be tailored for each OS. That is easy to do given a working executable.

I can do this given that you have setup a simple CLI.

andreped commented 1 year ago

This issue is linked to: https://github.com/jpdefrutos/DDMR/issues/11

I don't see it is critical to share the liver model, but it would be great to share the brain model, as it is more relevant for finetuning applications.

andreped commented 12 months ago

@jpdefrutos Shall we take a look at this October 2nd?

We both have more urgent deadlines in September, but I think it is time that we address this or the other issues.