Number of parameters in the nnUNet

[mchlsdrv]

Hi, I'd like to ask for the number of tunable parameters present in the network? Is it the same as the regular UNet?

Thanks in advance.
Michael

[FabianIsensee]

Hi,
honestly I don't know but there are tools that can tell you the number of parameters in pytorch models. That is what I would use.
Best,
Fabian

[Peaceandmaths]

I run this code and it worked : found 31195594 parameters ( but for some reason trainable and total parameters is the same number)

import torch from nnunetv2.utilities.get_network_from_plans import get_network_from_plans from nnunetv2.utilities.plans_handling.plans_handler import PlansManager from nnunetv2.training.nnUNetTrainer.nnUNetTrainer import nnUNetTrainer from nnunetv2.paths import nnUNet_results from batchgenerators.utilities.file_and_folder_operations import load_json from nnunetv2.paths import nnUNet_preprocessed, nnUNet_results from batchgenerators.utilities.file_and_folder_operations import join

# Replace with actual paths and configuration names as per your setup plans_path = '/path/to/nnUNetPlans.json' dataset_json_path = '/path/to/dataset.json' configuration = '3d_fullres' # Example configuration, replace with your actual configuration fold = 0 # Example fold

# Load plans and dataset_json plans = load_json(plans_path) dataset_json = load_json(dataset_json_path)

# Initialize the PlansManager and ConfigurationManager plans_manager = PlansManager(plans) configuration_manager = plans_manager.get_configuration(configuration)

# Ensure nnUNet_preprocessed and nnUNet_results are not None if nnUNet_preprocessed is None or nnUNet_results is None: raise ValueError("nnUNet_preprocessed and nnUNet_results must be set to valid paths.")

# Initialize the trainer
trainer = nnUNetTrainer(plans, configuration, fold, dataset_json)
# Initialize the network trainer.initialize()

# Function to count parameters def count_parameters(model): total_params = sum(p.numel() for p in model.parameters()) trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad) return total_params, trainable_params

# Get the number of parameters total_params, trainable_params = count_parameters(trainer.network) print(f'Total parameters in nnUNet model: {total_params}') print(f'Trainable parameters in nnUNet model: {trainable_params}')