# Load liver crops info
dict_file = liver_crops_dir.replace('ct', 'seg') + '/crop_list.p'
try:
import cPickle as pickle
except ImportError: # python 3.x
import pickle
with open(dict_file, 'rb') as fp:
crop_dict = pickle.load(fp)
masks_crop_dir = liver_crops_dir.replace('ct', 'seg')
# Load model
with tf.device(DEVICE):
# Get Lesion model
print('Getting lesion model...')
model = get_model(Config, inference=True)
print('loading weights...')
model.load_weights(weights_path)
print('Done!')
# Local Only
dice_lesion = np.zeros(num_files)
precision_lesion = np.zeros(num_files)
sensitivity_lesion = np.zeros(num_files)
print('len val_filenames: ', num_files)
print('num patients: ', num_patients)
n = 0
a = Progbar(num_files)
for i in range(num_patients):
masks_path = Config.labels_path
train_filenames, val_filenames = utils.split_filenames_train_val(data_path, val_prec=val_prec)
# Split validation and training paths
train_img_paths = [os.path.join(data_path, filename) for filename in train_filenames]
train_mask_paths = [os.path.join(masks_path, filename.replace('ct', 'seg')) for filename in train_filenames]
val_img_paths = [os.path.join(data_path, filename) for filename in val_filenames]
val_mask_paths = [os.path.join(masks_path, filename.replace('ct', 'seg')) for filename in val_filenames]
print('training patients indices: ', utils.get_unique_indices(train_filenames))
print('validation patients indices: ', utils.get_unique_indices(val_filenames))
# Get model
print('\ngetting lesion model...')
model_lesion = get_model(Config, freeze_encoder=freeze_encoder)
if load_model_weights:
print('loading lesion weights')
model_lesion.load_weights(load_model_weights)
print('\nDone!\n')
# init data generator
dc = utils.DataClass()
image_datagen = myDataGenerator(rotation_range=15,
zoom_range=0.1,
horizontal_flip=True,
height_shift_range=0.05,
width_shift_range=0.05,
data_format='channels_last',
subtract_mean=Config.mean,
masks_path = Config.labels_path
train_filenames, val_filenames = utils.split_filenames_train_val(data_path, val_prec=val_prec)
# Split validation and training paths
train_img_paths = [os.path.join(data_path, filename) for filename in train_filenames]
train_mask_paths = [os.path.join(masks_path, filename.replace('ct', 'seg')) for filename in train_filenames]
val_img_paths = [os.path.join(data_path, filename) for filename in val_filenames]
val_mask_paths = [os.path.join(masks_path, filename.replace('ct', 'seg')) for filename in val_filenames]
print('training patients indices: ', utils.get_unique_indices(train_filenames))
print('validation patients indices: ', utils.get_unique_indices(val_filenames))
# Get model
print('\ngetting model...')
model = get_model(Config, encoder_weights=encoder_weights)
if load_model_weights:
print('loading weight:', load_model_weights)
model.load_weights(load_model_weights, by_name=True)
print('\nDone!\n')
# init data generator
dc = utils.DataClass()
image_datagen = myDataGenerator(rotation_range=15,
zoom_range=0.1,
horizontal_flip=False,
height_shift_range=0.05,
width_shift_range=0.05,
data_format='channels_last',
subtract_mean=Config.mean,
crop_dict = {}
# Load data
print('Running liver detection on data!!!!!\n')
filenames, _ = utils.split_filenames_train_val(data_path, val_prec=0)
all_filenames_split = utils.split_to_patients(filenames)
# Display data info
num_patients = len(all_filenames_split)
num_files = sum([len(x) for x in all_filenames_split])
print('number of samples: ', num_files)
print('num patients: ', num_patients)
# Load liver segmentation model + weights
print('\ngetting liver model...')
model_liver = get_model(Config, inference=True)
print('loading liver weight:', weights_path)
model_liver.load_weights(weights_path)
print('\nDone!\n')
## Run model & extract liver ROIs
#####################
n = 0
a = Progbar(num_files)
dc = utils.DataClass()
for i in range(num_patients):
# load data to array
img_filenames = all_filenames_split[i]
mask_arr = np.zeros((len(img_filenames), orig_width, orig_height))
img_arr = np.zeros(