Refactor CLI for micro-sam backend support

[anwai98]

WIP!

[anwai98]

Okay this is sorted now! @constantinpape

Some details on how things would work:

1. Needs a pip install -e . in the current branch
2. Check out some simple integrations, eg. micro_sam --help / micro_sam --version / micro_sam info
3. And the other ports:
   a. micro_sam.annotator -> micro_sam segmentation_annotator
   b. micro_sam.annotator_tracking -> micro_sam tracking_annotator
   c. micro_sam.automatic_segmentation (SAM1) -> micro_sam v1 automatic_segmentation
   d. micro_sam.train (SAM1) -> micro_sam v1 train
   e. micro_sam.evaluate / micro_sam.benchmark_sam -> micro_sam v1 evaluate / micro_sam v1 benchmark_sam
   f. micro_sam.precompute_embeddings / micro_sam.info -> micro_sam precompute_embeddings / micro_sam info
   g. ...and all other CLIs too

[constantinpape]

I will test this properly tomorrow morning. Noticed one minor thing that could be improved in the code review.

[constantinpape]

Here's some more feedback:

General:

• Shorten the first part of the doc string so that it is not cut from the general help message (see screenshot below)
• It would be good to have a short usage example in the help message of each command (except info and v1).
• Unify naming "-i" -> "--input" (currently varies between input, input_path, input_folder); "-o" -> "--outtput"
• Rename halo -> overlap (mirror change in GUI)

Specific:

• automatic_segmentation:
  • add an optional --view to display results in napari?
  • segmentation in a folder does not work: micro_sam automatic_segmentation --input_path /home/pape/.cache/micro_sam/sample_data/image-series.zip.unzip/series/ -o test-out-seg --pattern *.tif fails with File "/home/pape/Work/my_projects/micro-sam/micro_sam/cli.py", line 324, in automatic_segmentation assert len(input_paths) > 0, "'micro-sam' could not extract any image data internally."; folder contains 3 tif images
  • Note: it actually works when passing "*.tif", so probably just an issue in argument parsing
• batch_annotator: opens the segmentation annotator but should support all of the annotators

I haven't checked the rest in detail

[constantinpape]

I think we should use the SAM2 embedding function by default --- and use our uSAM2 model as the default model.

[anwai98]

I see. Then I'll wrap this out around it like that!

[constantinpape]

Let's address this separately.

[constantinpape]

Should we merge this @anwai98 ?

[anwai98]

Oh I kind of was waiting for your approval here haha. If you assessment is positive here, let's go for it. It is GTM from my side!

[constantinpape]

Yeah let's merge it.

[anwai98]

Nice! ;)