Bugs when testing pretrained CartoCell BMZ models

[pedgomgal1]

Hi everyone,

I would like to report several issues I encountered while applying BioImage Model Zoo (BMZ) models in BiaPy. I hope this feedback helps improve the tool further.

First, I had to modify some automatically generated parameters within the .yaml when loading the BMZ models in BiaPy.

1. Output channels were BC. I had to M as for CartoCell to align with the remaining parameters.

2. I tried the "venomous-swan" model, and in the .yaml was automatically assigned as:

MODEL:
  BMZ:
    SOURCE_MODEL_ID: venomous-swan (BioImage Model Zoo)

Biapy crashes if I don´t remove manually " (BioImage Model Zoo)" as follows:

MODEL:
  BMZ:
    SOURCE_MODEL_ID: venomous-swan

After applying the two adjustments above, inference starts correctly and runs until the model-building step. At that point, BiaPy crashes with the following error:

100%|██████████| 60/60 [00:07<00:00,  7.58it/s]
100%|██████████| 60/60 [00:07<00:00,  8.32it/s]
Traceback (most recent call last):
  File "/usr/local/lib/python3.11/dist-packages/biapy/engine/base_workflow.py", line 779, in prepare_model
[10:31:14.928383] DATA.TEST.GT_PATH changed from /home/jesus/Escritorio/Pedro/CartoCell/test/y to /home/jesus/Escritorio/Pedro/CartoCell/test/y_F.erosion-0.dilation-0_C.mode-thick_M
[10:31:14.928684] ######################
[10:31:14.928689] #   LOAD TEST DATA   #
[10:31:14.928691] ######################
[10:31:14.928699] ### LOAD ###
[10:31:14.929233] ###############
[10:31:14.929237] # Build model #
    self.bmz_pipeline = create_prediction_pipeline(
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/bioimageio/core/_prediction_pipeline.py", line 371, in create_prediction_pipeline
[10:31:14.929239] ###############
[10:31:14.929271] The error thrown during the BMZ model load was:
expected v0_4.ModelDescr or v0_5.ModelDescr, but got <class 'bioimageio.spec._internal.common_nodes.InvalidDescr'>
    model_adapter = model_adapter or create_model_adapter(
                                     ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/bioimageio/core/backends/_model_adapter.py", line 86, in create
    raise TypeError(
TypeError: expected v0_4.ModelDescr or v0_5.ModelDescr, but got <class 'bioimageio.spec._internal.common_nodes.InvalidDescr'>

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/local/bin/biapy", line 6, in <module>
    sys.exit(main())
             ^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/biapy/__init__.py", line 141, in main
    _biapy.run_job()
  File "/usr/local/lib/python3.11/dist-packages/biapy/_biapy.py", line 1138, in run_job
    self.test()
  File "/usr/local/lib/python3.11/dist-packages/biapy/_biapy.py", line 260, in test
    self.workflow.test()
  File "/usr/local/lib/python3.11/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/dist-packages/biapy/engine/base_workflow.py", line 1219, in test
    self.prepare_model()
  File "/usr/local/lib/python3.11/dist-packages/biapy/engine/base_workflow.py", line 786, in prepare_model
    raise ValueError(
ValueError: An error ocurred when creating the BMZ model (see above). BiaPy only supports models prepared with pytorch_state_dict.

This happens during the model creation step, after test data loading and just before inference.

Sorry for the number of issue reports today — it’s been an intense debugging day. I hope this information is useful and helps improve BiaPy even further.

Best wishes,
BW