def test_generators_shift_in_flow(ct_path):
ct_array, meta = load_ct.load_ct(ct_path)
batch = np.expand_dims(np.expand_dims(ct_array[45:90], 0), 0)
dg = generators.DataGenerator(shift_range=[.1, .4, .4],
data_format='channels_first')
augmented = next(dg.flow(batch, batch_size=1, seed=21))
assert len(augmented.shape) == 5
assert augmented.shape == batch.shape
np.random.seed(21)
shifted = generators.random_shift(batch[0], (.1, .4, .4))
assert not np.abs(shifted - augmented).sum()
def set_params(self):
self.clip_lower = -1000
self.clip_upper = 400
self.train_data_generator = generators.DataGenerator(
rotation_range=20,
shift_range=.1,
zoom_lower=.8,
zoom_upper=1.2,
zoom_independent=False,
fill_mode='nearest',
preprocessing_function=None)
self.test_data_generator = generators.DataGenerator()
if self.data_format is None:
self.data_format = keras.backend.image_data_format()
if self.data_format == 'channels_last':
self.channel_axis = 4
elif self.data_format == 'channel_first':
self.channel_axis = 1
else:
raise ValueError(
"The `data_format` should lie in {None, 'channels_last', 'channels_first'}."
)
def test_generators_zoom_in_flow(ct_path):
ct_array, meta = load_ct.load_ct(ct_path)
batch = np.expand_dims(np.expand_dims(ct_array[45:90], 0), 0)
dg = generators.DataGenerator(zoom_lower=[1, .8, .8],
zoom_upper=[1, 1.2, 1.2],
zoom_independent=True,
data_format='channels_first')
augmented = next(dg.flow(batch, batch_size=1, seed=21))
assert len(augmented.shape) == 5
assert augmented.shape == batch.shape
np.random.seed(21)
zoomed = generators.random_zoom(batch[0],
zoom_lower=[1, .8, .8],
zoom_upper=[1, 1.2, 1.2],
independent=True)
assert not np.abs(zoomed - augmented).sum()
def test_generators_standardization(ct_path):
ct_array, meta = load_ct.load_ct(ct_path)
batch = np.expand_dims(np.expand_dims(ct_array[45:90], 0), 0)
dg = generators.DataGenerator(featurewise_center=True,
featurewise_std_normalization=True,
samplewise_center=True,
samplewise_std_normalization=True)
with warnings.catch_warnings():
warnings.filterwarnings('error')
try:
next(dg.flow(batch, batch_size=1))
except Warning as w:
assert "Fit it first by calling `.fit(numpy_data)`" in str(w)
dg.fit(batch)
normalized = next(dg.flow(batch, batch_size=1))
assert len(normalized.shape) == 5
assert normalized.shape == batch.shape
def test_generators_save_to_dir(ct_path, tmpdir):
ct_array, meta = load_ct.load_ct(ct_path)
batch = np.expand_dims(np.expand_dims(ct_array[45:90], 0), 0)
dg = generators.DataGenerator()
np.random.seed(1)
save_prefix = 'test'
fname = '{prefix}_{index}_{hash}.{format}'.format(
prefix=save_prefix, index=0, hash=np.random.randint(1e4), format='npy')
fname = str(tmpdir.mkdir("processed").join(fname))
batch = next(
dg.flow(batch,
seed=1,
batch_size=1,
save_to_dir=os.path.dirname(fname),
save_prefix=save_prefix))
saved = np.load(fname)
assert len(saved.shape) == 4
assert saved.shape == batch[0].shape
assert not np.abs(saved - batch[0]).sum()
