predict_nnunet.py

import torch
from nnunetv2.inference.predict_from_raw_data import nnUNetPredictor

def predictNNUNet(model_dir, input_dir, output_dir, folds):
    
    predictor = nnUNetPredictor(
        tile_step_size=0.9, #0.5,
        use_gaussian=True,
        use_mirroring=False,  # --disable_tta
        # perform_everything_on_device=True,
        device=torch.device('cpu', 0),
        verbose=True,
        verbose_preprocessing=False,
        allow_tqdm=True,
    )

    predictor.initialize_from_trained_model_folder(
        model_dir,
        use_folds=folds, # None if autodetect folds
        checkpoint_name='checkpoint_final.pth',
    )
    print("input_dir",input_dir)
    predictor.predict_from_files(input_dir,
                                 output_dir,
                                 save_probabilities=False,
                                 overwrite=True,
                                 num_processes_preprocessing=2,
                                 num_processes_segmentation_export=2,
                                 folder_with_segs_from_prev_stage=None,
                                 num_parts=1,
                                 part_id=0
                                 )