def testUnknownShape(self):
if tf.executing_eagerly():
# Placeholders don't work in eager execution mode.
return
inputs = tf.compat.v1.placeholder(tf.float32)
with self.assertRaisesRegexp(ValueError, 'undefined rank'):
norm.instance_norm(inputs)
def testParamsShapeNotFullyDefinedNHWC(self):
if tf.executing_eagerly():
# Placeholders don't work in eager execution mode.
return
inputs = tf.compat.v1.placeholder(tf.float32, shape=(3, 4, None))
with self.assertRaisesRegexp(ValueError, 'undefined channels dimension'):
norm.instance_norm(inputs, data_format='NHWC')
def testCreateVariables(self):
if tf.executing_eagerly():
# Collections don't work with eager.
return
height, width = 3, 3
images = tf.random.uniform((5, height, width, 3), seed=1)
norm.instance_norm(images, center=True, scale=True)
self.assertLen(contrib_utils.get_variables_by_name('beta'), 1)
self.assertLen(contrib_utils.get_variables_by_name('gamma'), 1)
def testReuseVariables(self):
if tf.executing_eagerly():
# Variable reuse doesn't work with eager.
return
height, width = 3, 3
images = tf.random.uniform((5, height, width, 3), seed=1)
norm.instance_norm(images, scale=True, scope='IN')
norm.instance_norm(images, scale=True, scope='IN', reuse=True)
self.assertLen(contrib_utils.get_variables_by_name('beta'), 1)
self.assertLen(contrib_utils.get_variables_by_name('gamma'), 1)
def testValueCorrectWithReuseVars(self):
height, width = 3, 3
image_shape = (10, height, width, 3)
images = tf.random.uniform(image_shape, seed=1)
output_train = norm.instance_norm(images, scope='IN')
output_eval = norm.instance_norm(images, scope='IN', reuse=True)
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
# output_train and output_eval should be the same.
train_np, eval_np = sess.run([output_train, output_eval])
self.assertAllClose(train_np, eval_np)
def testCreateOpFloat64(self):
height, width = 3, 3
images = tf.random.uniform((5, height, width, 3),
dtype=tf.float64,
seed=1)
output = norm.instance_norm(images)
self.assertListEqual([5, height, width, 3], output.shape.as_list())
def doOutputTest(self, input_shape, data_format, tol=1e-3):
axis = -1 if data_format == 'NHWC' else 1
for mu in (0.0, 1e2):
for sigma in (1.0, 0.1):
# Determine shape of Tensor after norm.
reduced_shape = (input_shape[0], input_shape[axis])
expected_mean = np.zeros(reduced_shape)
expected_var = np.ones(reduced_shape)
# Determine axes that will be normalized.
reduced_axes = list(range(len(input_shape)))
del reduced_axes[axis]
del reduced_axes[0]
reduced_axes = tuple(reduced_axes)
inputs = tf.random.uniform(input_shape, seed=0) * sigma + mu
output_op = norm.instance_norm(inputs,
center=False,
scale=False,
data_format=data_format)
with self.cached_session() as sess:
sess.run(tf.compat.v1.global_variables_initializer())
outputs = sess.run(output_op)
# Make sure that there are no NaNs
self.assertFalse(np.isnan(outputs).any())
mean = np.mean(outputs, axis=reduced_axes)
var = np.var(outputs, axis=reduced_axes)
# The mean and variance of each example should be close to 0 and 1
# respectively.
self.assertAllClose(expected_mean,
mean,
rtol=tol,
atol=tol)
self.assertAllClose(expected_var, var, rtol=tol, atol=tol)
def testCreateOpNoScaleCenter(self):
if tf.executing_eagerly():
# Collections don't work with eager.
return
height, width = 3, 3
images = tf.random.uniform((5, height, width, 3), dtype=tf.float64, seed=1)
output = norm.instance_norm(images, center=False, scale=False)
self.assertListEqual([5, height, width, 3], output.shape.as_list())
self.assertEmpty(contrib_utils.get_variables_by_name('beta'))
self.assertEmpty(contrib_utils.get_variables_by_name('gamma'))
def testBadDataFormat(self):
inputs = tf.zeros((2, 5, 5), dtype=tf.float32)
with self.assertRaisesRegexp(
ValueError, 'data_format has to be either NCHW or NHWC.'):
norm.instance_norm(inputs, data_format='NHCW')
