Unable to train model

[rvanroey]

Dear Kevin,

first of all thanks for making such an amazing tool and all the hard work in Omnipose. I've been trying to switch from cellpose to omnipose and followed the documentation on a Windows PC and Macbook to install the Omnipose GUI model. Although I am able to run the models that are incorporated, I can never train a new model with masks (its not clickable). In cellpose I never had this problem and now the issue persists on two different operation systems. Is this problem known or do you perhaps know how to fix it?

[kevinjohncutler]

@rvanroey Sorry that this feature is missing. I never played around with training using the GUI. I will put that on my list, but I'm not sure when I can get to it. In the meantime, I am happy to help you set up training using the CLI.

[rvanroey]

@kevinjohncutler, no apologies necessary. I'm already extremely grateful that you developed this tool; without it, my project would have experienced major delays. Initially, I believed that this feature was unavailable due to my own mistake. Personally, I am able to train the model using the CLI, but I thought it would be easier for my colleagues to use the GUI for training. I will take the time to show them how to use it effectively.

Regarding the GUI and creating masks (I particularly appreciate the various LUTs available), I wanted to bring to your attention an occasional error that occurs when drawing new masks. The error message is as follows: "QGraphicsScene::removeItem: item 0x2000cadd1d0's scene (0x0) is different from this scene (0x200036f7740)." Are you aware of this issue?

[kevinjohncutler]

@rvanroey That makes sense. A little expertise with the CLI will open up a whole new world of capabilities for them in the meantime!

Thank you for the bug report. I have not seen that error, but I'll look out for it when I use the GUI.

[JB4Jaison]

I'm having the same issue of training a custom model using the GUI. The images and the corresponding .seg label files are present in the folder but the option to train using them is disabled in the GUI.

These are the steps I followed for building my mamba environment on my Windows 11 machine:

mamba create -n omnipose 'python==3.9.18' pytorch torchvision pytorch-cuda=11.8 -c pytorch -c nvidia -y
mamba activate omnipose
git clone https://github.com/kevinjohncutler/omnipose.git
cd omnipose
pip install -e .

The version of Omnipose that I have is: 1.0.7.dev23+ga585929

Happy to share more details to clarify my situation.