def load_train_data(filepath, split=split, random_state=random_state):
id, features_train, target_train = load_train(filepath)
image_train = load_image(id)
sss = StratifiedShuffleSplit(n_splits=1,
train_size=split,
random_state=random_state)
train_i, val_i = next(sss.split(features_train, target_train))
features_val, val_img, val_target = features_train[val_i], image_train[
val_i], target_train[val_i]
features_train, train_img, train_target = features_train[
train_i], image_train[train_i], target_train[train_i]
return (features_train, train_img, train_target), (features_val, val_img,
val_target)
def main():
args = parser.parse_args()
learning_rate = args.learning_rate
if args.data_load == 'True':
images = np.load('images.npy')
labels = np.load('labels.npy')
elif args.data_load == 'False':
dataset = load_image(args.image_dir, args.n_process, args.shape)
images, labels = dataset
np.save('images.npy', images)
np.save('labels.npy', labels)
else:
print('Data load 인자 확인요망')
return
print('Dataset Shape :', images.shape)
counter = 1
epoch = 1
counter_path = os.path.join(CHECKPOINT_DIR, args.run_name, 'counter.json')
if os.path.exists(counter_path):
with open(counter_path, 'r') as f:
ce = json.load(f)
counter = ce['counter']
epoch = ce['epoch']
model_path = os.path.join(CHECKPOINT_DIR, args.run_name, 'best_model.h5')
mckpt = ModelCheckpoint(filepath=model_path,
monitor='val_loss',
verbose=1,
save_best_only=True)
input_image = Input(shape=args.shape + (1, ))
c_ae = model(args, input_image)
c_ae.compile(optimizer=Adam(lr=args.learning_rate),
loss='binary_crossentropy')
c_ae.summary()
# images = images / 255.
c_ae.fit(images,
images,
batch_size=args.batch_size,
epochs=args.epochs,
shuffle=True,
validation_split=0.2,
callbacks=[mckpt, SampleAndReconstruct(args)])
def load_test_data(filepath):
id, test_features = load_test(filepath)
test_images = load_image(id)
return (id, test_features, test_images)
def main():
images = load_image(110)
find_normal_vector_edges_like_paper(images[0])
import numpy as np
import load_images
def calculate_possible_values(image):
missing_values = np.ma.masked_equal(0, image)
print(missing_values)
quit()
if __name__ == '__main__':
image = load_images.load_image(0)
calculate_possible_values(image)
quit()