Train nnUNet with custom patch size

[noisyPixel]

I want to train the nnUNet with a custom input size of say 500 x 500. Is changing the plans.json enough after the preprocessing is complete?

[noisyPixel]

after the nnUNetPlans.json is created, i edited the 2d configuration to take a patch size of 400 x 400 instead of 448 x 384. but i see that during the forward pass, there is a dimension mismatch in the upsampling layer. so i am assuming there is some other place where we need to specify the patch size.

[noisyPixel]

I made this modification to the "\nnUNet_preprocessed\Dataset001_ABC\nnUNetPlans.json" and preprocessed and tried to train the model with this configuration.
"2d_patch400": {
"inherits_from": "2d",
"patch_size": [
400,
400
]
}