options['training_samples'] = curr_train_patients
options['test_samples'] = [curr_test_patient]
experiment_folder = jp(path_res, experiment_name)
create_folder(experiment_folder)
path_results = jp(experiment_folder, "fold" + str(fold).zfill(2))
create_folder(path_results)
path_models = jp(path_results, "models")
create_folder(path_models)
path_segmentations = jp(path_results, "results")
create_folder(path_segmentations)
# Organize the data in a dictionary
input_dictionary = create_training_validation_sets(options,
dataset_mode="l")
input_dictionary = expand_dictionary(input_dictionary)
transf = transforms.ToTensor()
print('Training data: ')
training_dataset = SlicesLoaderLoadAll(
input_data=input_dictionary['input_train_data'],
labels=input_dictionary['input_train_labels'],
roi=input_dictionary['input_train_rois'],
normalize=options['normalize'],
norm_type=options['norm_type'])
hola = training_dataset.__getitem__(100)
options['normalize'] = True
options['norm_type'] = 'zero_one'
options['batch_size'] = 10
options['patience'] = 20 #Patience for the early stopping
options['gpu_use'] = True
options['num_epochs'] = 200
options['optimizer'] = 'adam'
# Patch sampling type
#options['patch_sampling'] = 'mask' # (mask, balanced or balanced+roi or non-uniform)
# Loss
options['loss'] = 'dice' # (dice, cross-entropy)
# Whether or not to re-sample each epoch for training patches
options['resample_each_epoch'] = False
# Organize the data in a dictionary
input_dictionary = create_training_validation_sets(options)
#Show one subject with brain mask
#case_to_show = list(input_dictionary['input_val_data'].keys())[0] #First case of validation set
#shim_overlay(nib.load(input_dictionary['input_val_data'][case_to_show][0]).get_fdata(), nib.load(input_dictionary['input_val_rois'][case_to_show][0]).get_fdata(), 16, alpha=0.5)
# Create training, validation and test patches
transf = transforms.ToTensor()
"""
print('Training data: ')
training_dataset = SlicesLoader(input_data=input_dictionary['input_train_data'],
labels=input_dictionary['input_train_labels'],
roi=input_dictionary['input_train_rois'],
normalize=options['normalize'],
options['training_samples'] = curr_train_patients
options['test_samples'] = [curr_test_patient]
experiment_folder = jp(path_res, experiment_name)
create_folder(experiment_folder)
path_results = jp(experiment_folder, "fold" + str(fold).zfill(2))
create_folder(path_results)
path_models = jp(path_results, "models")
create_folder(path_models)
path_segmentations = jp(path_results, "results")
create_folder(path_segmentations)
# Organize the data in a dictionary
input_dictionary = create_training_validation_sets(
options, dataset_mode="l", specific_val=[validation_images[fold - 1]])
# Create training, validation and test patches
rotation_angle = 5
transf = transforms.Compose([
RandomFlipX(),
RandomFlipY(),
RandomFlipZ(),
#RandomRotationXY(rotation_angle),
#RandomRotationYZ(rotation_angle),
#RandomRotationXZ(rotation_angle),
ToTensor3DPatch()
])
print('Training data: ')