config.py

from dataset.lidc import LIDC
from dataset.lungx import LUNGx

class Config():
    def __init__(self):
        super(Config, self).__init__()


def load_config():
      
    cfg = Config()
    ''' Experiment '''
    cfg.experiment_idx = 1 
    cfg.trial_id = None

    cfg.device = "cuda:0"
    cfg.save_dir_prefix = 'Experiment_' # prefix for experiment folder
    cfg.name = 'voxel2mesh'

    ''' 
    **************************************** Paths ****************************************
    save_path: results will be saved at this location
    dataset_path: dataset must be stored here.
    '''
    cfg.save_path = './experiments/MICCAI2022/'   # UPDATE HERE <<<<<<<<<<<<<<<<<<<<<<
    cfg.dataset_path = './DATA/LIDC_spiculation/generated_vol_iso64x64x64/' # UPDATE HERE <<<<<<<<<<<<<<<<<<<<<<
    cfg.ext_dataset_path = './DATA/LUNGx_spiculation/generated_vol_iso64x64x64/' # UPDATE HERE <<<<<<<<<<<<<<<<<<<<<<
    
    # cfg.save_path = '/your/path/to/experiments/MICCAI2022/' # results will be saved here
    # cfg.dataset_path = '/your/path/to/dataset' # path to the dataset
    # cfg.ext_dataset_path = '/your/path/to/dataset' # path to the dataset for external validation

    # Initialize data object for. 
    # LIDC() for LIDC-IDRI and LUNGx() for LUNGx dataset. 

    #cfg.data_obj = None     # UPDATE HERE <<<<<<<<<<<<<<<<<<<<<<
    cfg.data_obj = LIDC() 
    cfg.data_obj_ext = LUNGx() 


    assert cfg.save_path != None, "Set cfg.save_path in config.py"
    assert cfg.dataset_path != None, "Set cfg.dataset_path in config.py"
    assert cfg.data_obj != None, "Set cfg.data_obj in config.py"

    ''' 
    ************************************************************************************************
    ''' 




    ''' Dataset '''  
    # input should be cubic. Otherwise, input should be padded accordingly.
    cfg.patch_shape = (64, 64, 64) 
    

    cfg.ndims = 3
    cfg.augmentation_shift_range = 10
    
    ''' Malignancy Classifier '''
    # Set to True to use encoder features in addition to mesh features for
    # Malignancy classification
    cfg.deep_features_classifier = False #True
    
    ''' Model '''
    cfg.first_layer_channels = 16
    cfg.num_input_channels = 1
    cfg.steps = 4

    # Only supports batch size 1 at the moment. 
    cfg.batch_size = 1


    cfg.num_classes = 4
    cfg.batch_norm = False  
    cfg.graph_conv_layer_count = 4

  
    ''' Optimizer '''
    cfg.learning_rate = 1e-4
    cfg.weight_decay = 1e-7

    ''' Training '''
    cfg.numb_of_epochs = 200

    ''' Rreporting '''
    cfg.wab = True # use weight and biases for reporting
    
    return cfg