def testBadCommensurateGroup(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(5, 4, 10, 10))
with self.assertRaisesRegexp(ValueError,
'4 channels is not commensurate with '
'3 groups.'):
normalization.group_norm(inputs, groups=3,
reduction_axes=[-2, -1], channels_axis=-3)
def testParamsShapeNotFullyDefinedChannelsAxis(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(1, 3, 4, None))
with self.assertRaisesRegexp(ValueError,
'undefined channel dimension'):
normalization.group_norm(inputs,
channels_axis=-1,
reduction_axes=[-3, -2])
def testInvalidGroupSize(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(5, 2, 10, 10))
with self.assertRaisesRegexp(ValueError,
'Invalid groups 10 for 2 channels.'):
normalization.group_norm(inputs,
groups=10,
reduction_axes=[-2, -1],
channels_axis=-3)
def testAxisIsBad(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(1, 2, 4, 5))
with self.assertRaisesRegexp(ValueError,
'Axis is out of bounds.'):
normalization.group_norm(inputs, channels_axis=5)
with self.assertRaisesRegexp(ValueError,
'Axis is out of bounds.'):
normalization.group_norm(inputs, reduction_axes=[1, 5])
def testCreateVariables_NCHW(self):
height, width, groups = 3, 3, 4
images = random_ops.random_uniform((5, 2*groups, height, width), seed=1)
normalization.group_norm(images, groups=4,
channels_axis=-3, reduction_axes=(-2, -1),
center=True, scale=True)
beta = contrib_variables.get_variables_by_name('beta')[0]
gamma = contrib_variables.get_variables_by_name('gamma')[0]
self.assertEqual('GroupNorm/beta', beta.op.name)
self.assertEqual('GroupNorm/gamma', gamma.op.name)
def testReuseVariables(self):
height, width = 3, 3
images = random_ops.random_uniform((5, height, width, 4), seed=1)
normalization.group_norm(images, groups=2, scale=True, scope='IN')
normalization.group_norm(images, groups=2, scale=True, scope='IN',
reuse=True)
beta = contrib_variables.get_variables_by_name('beta')
gamma = contrib_variables.get_variables_by_name('gamma')
self.assertEqual(1, len(beta))
self.assertEqual(1, len(gamma))
def testValueCorrectWithReuseVars(self):
height, width = 3, 3
image_shape = (10, height, width, 4)
images = random_ops.random_uniform(image_shape, seed=1)
output_train = normalization.group_norm(images, groups=2, scope='IN')
output_eval = normalization.group_norm(images, groups=2, scope='IN',
reuse=True)
with self.test_session() as sess:
sess.run(variables.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 testCreateOp(self):
height, width, groups = 3, 3, 4
images = random_ops.random_uniform((5, height, width, 2*groups), seed=1)
output = normalization.group_norm(images, groups=groups, channels_axis=-1,
reduction_axes=[-3, -2])
print('name: ', output.op.name)
self.assertListEqual([5, height, width, 2*groups], output.shape.as_list())
def testCreateOpFloat64(self):
height, width, groups = 3, 3, 5
images = random_ops.random_uniform(
(5, height, width, 4*groups), dtype=dtypes.float64, seed=1)
output = normalization.group_norm(images, groups=groups)
self.assertEqual(dtypes.float64, output.dtype)
self.assertListEqual([5, height, width, 4*groups], output.shape.as_list())
def testParamsShapeNotFullyDefinedBatchAxis(self):
height, width, groups = 3, 3, 4
inputs = array_ops.placeholder(dtypes.float32,
shape=(None, height, width, 2*groups))
output = normalization.group_norm(inputs, channels_axis=-1,
reduction_axes=[-3, -2], groups=groups)
self.assertListEqual([None, height, width, 2 * groups],
output.shape.as_list())
def testCreateOpNoScaleCenter(self):
height, width, groups = 3, 3, 7
images = random_ops.random_uniform(
(5, height, width, 3*groups), dtype=dtypes.float32, seed=1)
output = normalization.group_norm(images, groups=groups, center=False,
scale=False)
self.assertListEqual([5, height, width, 3*groups], output.shape.as_list())
self.assertEqual(0, len(contrib_variables.get_variables_by_name('beta')))
self.assertEqual(0, len(contrib_variables.get_variables_by_name('gamma')))
def doOutputTest(self, input_shape, channels_axis=None, reduction_axes=None,
groups=2, tol=1e-2):
# Select the axis for the channel and the dimensions along which statistics
# are accumulated.
if channels_axis < 0:
channels_axis += len(input_shape)
reduced_axes = [channels_axis + 1]
for a in reduction_axes:
if a < 0:
a += len(input_shape)
if a < channels_axis:
reduced_axes.append(a)
else:
reduced_axes.append(a+1)
reduced_axes = tuple(reduced_axes)
# Calculate the final shape for the output Tensor.
axes_before_channels = input_shape[:channels_axis]
axes_after_channels = input_shape[channels_axis+1:]
channels = input_shape[channels_axis]
outputs_shape = (axes_before_channels + [groups, channels // groups] +
axes_after_channels)
# Calculate the final shape for the output statistics.
reduced_shape = []
for i, a in enumerate(outputs_shape):
if i not in reduced_axes:
reduced_shape.append(a)
for mu in (0.0, 1e2):
for sigma in (1.0, 0.1):
# Determine shape of Tensor after normalization.
expected_mean = np.zeros(reduced_shape)
expected_var = np.ones(reduced_shape)
inputs = random_ops.random_uniform(input_shape, seed=0) * sigma + mu
output_op = normalization.group_norm(
inputs, groups=groups, center=False, scale=False,
channels_axis=channels_axis,
reduction_axes=reduction_axes)
with self.test_session() as sess:
sess.run(variables.global_variables_initializer())
outputs = sess.run(output_op)
# Make sure that there are no NaNs
self.assertFalse(np.isnan(outputs).any())
outputs = np.reshape(outputs, outputs_shape)
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 testNotMutuallyExclusiveAxis(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(10, 32, 32, 32))
# Specify axis with negative values.
with self.assertRaisesRegexp(ValueError, 'mutually exclusive'):
normalization.group_norm(inputs, channels_axis=-2, reduction_axes=[-2])
# Specify axis with positive values.
with self.assertRaisesRegexp(ValueError, 'mutually exclusive'):
normalization.group_norm(inputs, channels_axis=1, reduction_axes=[1, 3])
# Specify axis with mixed positive and negative values.
with self.assertRaisesRegexp(ValueError, 'mutually exclusive'):
normalization.group_norm(inputs, channels_axis=-2, reduction_axes=[2])
def testParamsShapeNotFullyDefinedReductionAxes(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(1, 32, None, 4))
with self.assertRaisesRegexp(ValueError, 'undefined dimensions'):
normalization.group_norm(inputs)
def testInvalidGroupSize(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(5, 2, 10, 10))
with self.assertRaisesRegexp(ValueError,
'Invalid groups 10 for 2 channels.'):
normalization.group_norm(inputs, groups=10,
reduction_axes=[-2, -1], channels_axis=-3)
def testUnknownShape(self):
inputs = array_ops.placeholder(dtypes.float32)
with self.assertRaisesRegexp(ValueError, 'undefined rank'):
normalization.group_norm(inputs)
def testParamsShapeNotFullyDefinedChannelsAxis(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(1, 3, 4, None))
with self.assertRaisesRegexp(ValueError, 'undefined channel dimension'):
normalization.group_norm(inputs, channels_axis=-1,
reduction_axes=[-3, -2])
