Error in deep_watershed.py #134
Comments
|
Thanks for reporting @joshuagohos . Can you share how the environment was created (for example, did you use |
|
In Colab, I used Checking the scikit-image version in Colab, I get: 0.18.3 The package installation output for |
|
Interesting - that install log looks like a lot of old versions and unnecessary downloads. For example the cascading back through versions of numpy all the way back to 1.17 is probably dominating install time. Unfortunately it's always tough to say what causes things like this due to transitive dependencies. For this particular case, after you've run |
|
Ok. I've added it in the following: I am now able to run app.predict() based on the NuclearSegmentation application. So it works insofar as the previous bug is now gone. But, one caveat is that it seems to give me different segmentation results than what I got before when trying on my own image data. Before the above error and using the previous scikit-image version, applying nuclear-application segmentation gave me this segmentation: But with this update of scikit-image, the same app.predict() from. Nuclear Segmentation, I get: The color scaling rendering is a little different, but the segmentation is applied on the same base image. Moreover, both are different from what I get from the online DeepCell predict -> label result: I've reflected this in another issue intro-to-deepcell#45 |
rossbar
added a commit
to rossbar/deepcell-toolbox
that referenced
this issue
Dec 22, 2022
skimage-0.19 introduced the deprecations for the selem keyword args. This will fix issues like vanvalenlabgh-134.
Great, thanks for checking!
The scope here is much larger than the original issue. There are many potential sources of variability: model versions, data versions, any additions or alterations vis-a-vis model training (if applicable), pre- or post- processing steps that involve random processes, dependency management, etc. At this point I don't have any concrete suggestions for a fully reproducible workflow for your use-case. Some (non-expert) suggestions that I'd offer:
|
|
I just want to add that there was an update to the NuclearSegmentation model between deepcell 0.12.3 and deepcell 0.12.4 so I would second Ross's suggestion to revert to the previous version with |
|
Thanks. Concur with your suggestions and will try them in Colab. Are the Docker container versions more recommended then in this case? |
|
I confirm that reverting to deepcell==0.12.3 reinstates my previous segmentation results. Thanks! The only issue remaining is the comparability between the kiosk/online versions and the notebook application version. |
|
I checked and the same models should be deployed in both deepcell 0.12.3 and in the kiosk. We'll look into possible differences after the holidays and let you know if we find anything. |
|
Thanks. If it helps, I found that changing the resolution parameter in the notebook command modulates the segmentation result to be closer to the online version. In the online version, I set the image resolution to 1 micron to get the result. In the notebook, if I do app.predict(myimagearray, image_mpp=1.0), this yields the segmentation that is quite different. But if I do image_mpp = 1.3 the segmentation approaches the online one. Is the online version applying some additional rescaling beyond the initial 1 micron setting that the noteboook version does not do? In any case, happy holidays! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Running segmentations_predictions = app.predict(X_train, image_mpp=0.5) throws:
TypeError: h_maxima() got an unexpected keyword argument 'footprint'
Referencing /usr/local/lib/python3.8/dist-packages/deepcell_toolbox/deep_watershed.py
This was not present before (the commands worked fine). It just occurred around 12PM 21 Dec 2022. This same error also applies to the Nuclear-Application and Cytoplasm-Application.
This error occurs as I am trying out the notebook https://github.com/vanvalenlab/deepcell-tf/tree/master/notebooks/applications/Mesmer-Application.ipynb.
Here is the snapshow of the above ipynb as I ran it in Colab.
The text was updated successfully, but these errors were encountered: