Trying out the omnipose detector

[m-albert]

Hi @tinevez ,

here are some observations from trying out the omnipose detector using "TrackMate-Cellpose-0.1.3-SNAPSHOT.jar" on macos.

In a fresh python=3.10 conda environment:

1. pip install omnipose==0.3.6 (the documentation says the plugin is compatible with 0.3.6 but not with 0.4.4)

Results in "could not find results file".

When executing the command shown in a terminal (with omnipose activated) I get:

File "/Users/malbert/miniconda3/envs/omnipose/lib/python3.10/site-packages/omnipose/main.py", line 1, in
from cellpose.main import main
ModuleNotFoundError: No module named 'cellpose.main'

Indeed it seems that code in the wheel shipped over pypi tries to import cellpose instead of cellpose_omni.

2. pip install omnipose, which installed omnipose==0.4.4

It works 🎉

When segmenting images both with a pretrained model and a custom model, it shows sensible results:

I also tried for multichannel images and the field "channel to segment" seems to work properly (tried in combination with optional second channel: 0 none)