def forward(self, x):
with tf.variable_scope(self.Gscope):
h = self.downSampling1(x)
with tf.variable_scope("downSampling1_norm") as self.Gscope:
h = instance_norm(h, scale=False, epsilon=1e-5)
h = tf.nn.relu(h)
h = self.downSampling2(h)
with tf.variable_scope("downSampling2_norm") as self.Gscope:
h = instance_norm(h, scale=False, epsilon=1e-5)
h = tf.nn.relu(h)
h = self.downSampling3(h)
with tf.variable_scope("downSampling3_norm") as self.Gscope:
h = instance_norm(h, scale=False, epsilon=1e-5)
h = tf.nn.relu(h)
h = self.residualBlock1.forward(h)
h = self.residualBlock2.forward(h)
h = self.residualBlock3.forward(h)
h = self.residualBlock4.forward(h)
h = self.residualBlock5.forward(h)
h = self.residualBlock6.forward(h)
h = self.Y_upSampling1(h)
with tf.variable_scope("Y_upSampling1_norm") as self.Gscope:
h = instance_norm(h, scale=True, epsilon=1e-5)
h = tf.nn.relu(h)
h = self.Y_upSampling2(h)
with tf.variable_scope("Y_upSampling2_norm") as self.Gscope:
h = instance_norm(h, scale=True, epsilon=1e-5)
h = tf.nn.relu(h)
return self.fakeGeneration(h)
def testCreateVariables(self):
height, width = 3, 3
images = random_ops.random_uniform((5, height, width, 3), seed=1)
normalization.instance_norm(images, center=True, scale=True)
beta = contrib_variables.get_variables_by_name('beta')[0]
gamma = contrib_variables.get_variables_by_name('gamma')[0]
self.assertEqual('InstanceNorm/beta', beta.op.name)
self.assertEqual('InstanceNorm/gamma', gamma.op.name)
def testReuseVariables(self):
height, width = 3, 3
images = random_ops.random_uniform((5, height, width, 3), seed=1)
normalization.instance_norm(images, scale=True, scope='IN')
normalization.instance_norm(images, 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 forward(self, x):
h = self.Y1(x)
with tf.variable_scope("Generator/ResBlock" + str(self.num) + "1"):
h = instance_norm(h, scale=False, epsilon=1e-5)
h = tf.nn.relu(h)
h = self.Y2(h)
with tf.variable_scope("Generator/ResBlock" + str(self.num) + "2"):
h = instance_norm(h, scale=False, epsilon=1e-5)
return x + h
def testValueCorrectWithReuseVars(self):
height, width = 3, 3
image_shape = (10, height, width, 3)
images = random_ops.random_uniform(image_shape, seed=1)
output_train = normalization.instance_norm(images, scope='IN')
output_eval = normalization.instance_norm(images, 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 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 normalization.
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 = random_ops.random_uniform(input_shape, seed=0) * sigma + mu
output_op = normalization.instance_norm(
inputs, center=False, scale=False, data_format=data_format)
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())
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 testCreateOp(self):
height, width = 3, 3
images = random_ops.random_uniform((5, height, width, 3), seed=1)
output = normalization.instance_norm(images)
print('name: ', output.op.name)
self.assertStartsWith(output.op.name, 'InstanceNorm/instancenorm')
self.assertListEqual([5, height, width, 3], output.shape.as_list())
def testCreateOpFloat64(self):
height, width = 3, 3
images = random_ops.random_uniform(
(5, height, width, 3), dtype=dtypes.float64, seed=1)
output = normalization.instance_norm(images)
self.assertStartsWith(
output.op.name, 'InstanceNorm/instancenorm')
self.assertListEqual([5, height, width, 3], output.shape.as_list())
def testCreateOpNoScaleCenter(self):
height, width = 3, 3
images = random_ops.random_uniform(
(5, height, width, 3), dtype=dtypes.float64, seed=1)
output = normalization.instance_norm(images, center=False, scale=False)
self.assertStartsWith(
output.op.name, 'InstanceNorm/instancenorm')
self.assertListEqual([5, height, width, 3], 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 testParamsShapeNotFullyDefinedNHWC(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(3, 4, None))
with self.assertRaisesRegexp(ValueError, 'undefined channels dimension'):
normalization.instance_norm(inputs, data_format='NHWC')
def testBadDataFormat(self):
inputs = array_ops.placeholder(dtypes.float32, shape=(2, 5, 5))
with self.assertRaisesRegexp(ValueError,
'data_format has to be either NCHW or NHWC.'):
normalization.instance_norm(inputs, data_format='NHCW')
def testUnknownShape(self):
inputs = array_ops.placeholder(dtypes.float32)
with self.assertRaisesRegexp(ValueError, 'undefined rank'):
normalization.instance_norm(inputs)
def residual_block_with_IN(incoming,
nb_blocks,
out_channels,
downsample=False,
downsample_strides=2,
activation='relu',
batch_norm=True,
bias=True,
weights_init='variance_scaling',
bias_init='zeros',
regularizer='L2',
weight_decay=0.0001,
trainable=True,
restore=True,
reuse=False,
scope=None,
name="ResidualBlock",
is_training=True):
""" Residual Block.
A residual block as described in MSRA's Deep Residual Network paper.
Full pre-activation architecture is used here.
Input:
4-D Tensor [batch, height, width, in_channels].
Output:
4-D Tensor [batch, new height, new width, nb_filter].
Arguments:
incoming: `Tensor`. Incoming 4-D Layer.
nb_blocks: `int`. Number of layer blocks.
out_channels: `int`. The number of convolutional filters of the
convolution layers.
downsample: `bool`. If True, apply downsampling using
'downsample_strides' for strides.
downsample_strides: `int`. The strides to use when downsampling.
activation: `str` (name) or `function` (returning a `Tensor`).
Activation applied to this layer (see tflearn.activations).
Default: 'linear'.
batch_norm: `bool`. If True, apply batch normalization.
bias: `bool`. If True, a bias is used.
weights_init: `str` (name) or `Tensor`. Weights initialization.
(see tflearn.initializations) Default: 'uniform_scaling'.
bias_init: `str` (name) or `tf.Tensor`. Bias initialization.
(see tflearn.initializations) Default: 'zeros'.
regularizer: `str` (name) or `Tensor`. Add a regularizer to this
layer weights (see tflearn.regularizers). Default: None.
weight_decay: `float`. Regularizer decay parameter. Default: 0.001.
trainable: `bool`. If True, weights will be trainable.
restore: `bool`. If True, this layer weights will be restored when
loading a model.
reuse: `bool`. If True and 'scope' is provided, this layer variables
will be reused (shared).
scope: `str`. Define this layer scope (optional). A scope can be
used to share variables between layers. Note that scope will
override name.
name: A name for this layer (optional). Default: 'ShallowBottleneck'.
is_training: True for training mode and False for val or test mode.
References:
- Deep Residual Learning for Image Recognition. Kaiming He, Xiangyu
Zhang, Shaoqing Ren, Jian Sun. 2015.
- Identity Mappings in Deep Residual Networks. Kaiming He, Xiangyu
Zhang, Shaoqing Ren, Jian Sun. 2015.
Links:
- [http://arxiv.org/pdf/1512.03385v1.pdf]
(http://arxiv.org/pdf/1512.03385v1.pdf)
- [Identity Mappings in Deep Residual Networks]
(https://arxiv.org/pdf/1603.05027v2.pdf)
"""
resnet = incoming
in_channels = incoming.get_shape().as_list()[-1]
with tf.variable_scope(values=[incoming],
name_or_scope=scope,
default_name=name,
reuse=reuse) as scope:
name = scope.name # TODO
for i in range(nb_blocks):
identity = resnet
if not downsample:
downsample_strides = 1
if batch_norm:
resnet = normalization.instance_norm(resnet,
scope=name + '_bn_' +
str(i) + '_1')
resnet = tflearn.activation(resnet, activation)
resnet = tflearn.conv_2d(resnet, out_channels, 3,
downsample_strides, 'same', 'linear',
bias, weights_init, bias_init,
regularizer, weight_decay, trainable,
restore)
if batch_norm:
resnet = normalization.instance_norm(resnet,
scope=name + '_bn_' +
str(i) + '_2')
resnet = tflearn.activation(resnet, activation)
resnet = tflearn.conv_2d(resnet, out_channels, 3,
downsample_strides, 'same', 'linear',
bias, weights_init, bias_init,
regularizer, weight_decay, trainable,
restore)
# Downsampling
if downsample_strides > 1:
identity = tflearn.avg_pool_2d(identity, 1, downsample_strides)
# Projection to new dimension
'''
if in_channels != out_channels:
ch = (out_channels - in_channels)//2
identity = tf.pad(identity,
[[0, 0], [0, 0], [0, 0], [ch, ch]])
in_channels = out_channels
'''
resnet = resnet + identity
return resnet
