The Plans_handler.py does not match the original repo.

[Strawbili]

When I'm trying your models on other Dataset, I found the configManager went something wrong. There might be errors like AttributeError: 'ConfigurationManager' object has no attribute 'conv_kernel_sizes'. Did you mean: 'pool_op_kernel_sizes'?
I'm using the offical-generated nnUNet Plans Manager and I found your Plans Manager doesn't match the official ones. Did you modified the Plans_handler.py while you are coding? Why did you do this?

[wyd2]

I have the same problem. How can I solve it?

[Strawbili]

I've solved this problem by revise the code in the plans_handler.py.
To be more specific, please print/debug a variable called PlansManager.
Then alter the api as the structures in the PlansManager.
The name of those keys may vary a little bit from the original one:(

[Strawbili]

And I modified the code again, here are some hints below.
For instance, the modification of the code should be from:

    def conv_kernel_sizes(self) -> List[List[int]]:
        return self.configuration['conv_kernel_sizes']

to

    def conv_kernel_sizes(self) -> List[List[int]]:
        return self.configuration['architecture']['arch_kwargs']['kernel_sizes']

plz print the plans_manager variable in the function get_umamba_enc_2d_from_plans in the file umamba/nnunetv2/nets/UMambaEnc_2d.py

[wyd2]

Thank you so much for sharing your experience. However, I found that the error was caused by the incorrect version of nnunet. We should compile nnunet first to process the data, so that the generated plan.json is correct.

[hjj194]

I think the issue is that you're using nnUNet's plans and preprocess instead of UMamba's. This is likely because of some previous nnUNet setup that’s still active. Make sure you initialize the UMamba environment variables correctly. Also, you can use pip list to check which version of nnUNet is currently installed in your environment.