def load_custom_dataset(self, train_path, val_path):
"""
Loads the custom dataset from train_path and val_path.
"""
# Loads training dataset from train_path
with open(train_path, 'rb') as f:
load_dict = pickle.load(f)
self.y = load_dict['y']
self.x_true = load_dict['x_true']
self.fbp = load_dict['tv']
for idx in range(self.fbp.shape[0]):
self.fbp[idx, :, :, 0] = ut.scale_to_unit_intervall(self.fbp[idx, :, :, 0])
print("STARTING FROM TV")
print(f"self.fbp.max(): {self.fbp.max()}")
print(f"self.fbp.min(): {self.fbp.min()}")
self.batch_indices = np.arange(start=0, stop=self.x_true.shape[0], dtype=np.int)
np.random.shuffle(self.batch_indices)
# Loads validation dataset from validation path
with open(val_path, 'rb') as f:
val_load_dict = pickle.load(f)
self.val_y = val_load_dict['y']
self.val_x_true = val_load_dict['x_true']
self.val_fbp = val_load_dict['tv']
for idx in range(self.val_fbp.shape[0]):
self.val_fbp[idx, :, :, 0] = ut.scale_to_unit_intervall(self.val_fbp[idx, :, :, 0])
print("STARTING FROM TV")
print(f"self.val_fbp.max(): {self.val_fbp.max()}")
print(f"self.val_fbp.min(): {self.val_fbp.min()}")
self.val_size = self.val_y.shape[0]
def load_data(self, training_data= True):
output = np.zeros((128, 128, 1))
if training_data:
pic = self.random_phantom(spc= self.space)
output[..., 0] = ut.scale_to_unit_intervall(pic)
else:
pic = odl.phantom.shepp_logan(self.space, True)
output[..., 0] = ut.scale_to_unit_intervall(pic)
return output
def load_data(self, training_data=True):
pic = self.random_phantom(spc=self.space)
output = np.zeros((128, 128, 1))
output[..., 0] = ut.scale_to_unit_intervall(pic)
return output
def reshape_pic(self, pic):
pic = ut.normalize_image(pic)
pic = imresize(pic, [128, 128])
pic = ut.scale_to_unit_intervall(pic)
return pic
def reshape_pic(self, pic):
pic = ut.normalize_image(pic)
# pic = imresize(pic, [128, 128])
pic = np.array(Image.fromarray(pic).resize(size=(128, 128)))
pic = ut.scale_to_unit_intervall(pic)
return pic