Colab inference demo incompatible with current MONAI and NumPy

[msayr]

Hi there,

I was trying out the Colab inference notebook ([here])(https://colab.research.google.com/github/MMathisLab/CellSeg3d/blob/main/Notebooks/Colab_inference_demo.ipynb#scrollTo=IFbmZ3_zAm-1) and I've gotten the following error. All original code above this line ran without error:

result = cs3d.inference_on_images(
    demo_image,
    config=inference_config,
)

WARNING:napari_cellseg3d.utils:Caught TypeError: SwinUNETR.__init__() got multiple values for argument 'in_channels'
ERROR:napari_cellseg3d.utils:SwinUNETR.__init__() got multiple values for argument 'in_channels'
Traceback (most recent call last):
  File "/usr/local/lib/python3.12/dist-packages/napari_cellseg3d/code_models/models/model_SwinUNetR.py", line 34, in __init__
    super().__init__(
TypeError: SwinUNETR.__init__() got multiple values for argument 'in_channels'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/local/lib/python3.12/dist-packages/napari_cellseg3d/code_models/worker_inference.py", line 859, in inference
    model = model_class(
            ^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/napari_cellseg3d/code_models/models/model_SwinUNetR.py", line 47, in __init__
    super().__init__(
TypeError: SwinUNETR.__init__() got multiple values for argument 'in_channels'
--------------------
Parameters summary :
Model is : SwinUNetR
Window inference is enabled
Window size is 64
Window overlap is 0.25
Dataset loaded on cuda device
--------------------
MODEL DIMS : [64, 64, 64]
Model name : SwinUNetR
Instantiating model...

Changing numpy and monai versions solved the issue for me:

!pip install -U "numpy>=2.1,<2.6" "monai==1.3.2"

import numpy, monai
print("NumPy:", numpy.__version__)
print("MONAI:", monai.__version__)

NumPy: 2.4.6
MONAI: 1.3.2

Then updating reshape to shape to match different numpy version:

from pathlib import Path

p = Path("/usr/local/lib/python3.12/dist-packages/napari_cellseg3d/code_models/worker_inference.py")
s = p.read_text()
s = s.replace("np.reshape(volume, newshape=(1, *volume.shape))",
              "np.reshape(volume, (1, *volume.shape))")
p.write_text(s)

print("Patched:", p)

[C-Achard]

Hello @msayr,

Thanks a lot for reporting; I’m currently updating the package to run properly on Python 3.12 and to fix several deprecated/old calls, including the newshape issue.

The SwinUNETR.__init__() got multiple values for argument 'in_channels' error looks related to MONAI API/version compatibility, so I’m checking that as part of the update as well. In my limited local testing, the plugin is now running reasonably well with recent napari/MONAI versions, but I remember this being broken by a MONAI update, they might have fixed it in a subsequent release however as I got it to run locally. Will investigate this further.

Thanks again!

Best,
Cyril

[C-Achard]

Closed automatically by #120, re-opening until released/fix is confirmed