本文整理汇总了Python中tensorflow.contrib.slim.layer_norm方法的典型用法代码示例。如果您正苦于以下问题:Python slim.layer_norm方法的具体用法?Python slim.layer_norm怎么用?Python slim.layer_norm使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类tensorflow.contrib.slim
的用法示例。
在下文中一共展示了slim.layer_norm方法的12个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: argscope
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def argscope(is_training=None, normalizer_fn=slim.layer_norm):
"""Default TF argscope used for convnet-based grasping models.
Args:
is_training: Whether this argscope is for training or inference.
normalizer_fn: Which conv/fc normalizer to use.
Returns:
Dictionary of argument overrides.
"""
with slim.arg_scope([slim.batch_norm, slim.dropout], is_training=is_training):
with slim.arg_scope(
[slim.conv2d, slim.fully_connected],
weights_initializer=tf.truncated_normal_initializer(stddev=0.01),
activation_fn=tf.nn.relu,
normalizer_fn=normalizer_fn):
with slim.arg_scope(
[slim.conv2d, slim.max_pool2d], stride=2, padding='VALID') as scope:
return scope
示例2: __call__
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def __call__(self, inputs, state, scope=None):
if self._apply_to == 'input':
with tf.variable_scope(scope or self._name):
inputs = slim.layer_norm(inputs)
return self._cell(inputs, state)
elif self._apply_to == 'output':
output, res_state = self._cell(inputs, state)
with tf.variable_scope(scope or self._name):
output = slim.layer_norm(output)
return output, res_state
elif self._apply_to == 'state':
output, res_state = self._cell(inputs, state)
with tf.variable_scope(scope or self._name):
res_state = slim.layer_norm(res_state)
return output, res_state
else:
raise ValueError('Unknown apply_to: "{}"'.format(self._apply_to))
示例3: __call__
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def __call__(self, inputs, state, scope=None):
if self._apply_to == 'input':
with tf.variable_scope(scope or self._name):
inputs = slim.layer_norm(inputs)
return self._cell(inputs, state)
elif self._apply_to == 'output':
output, res_state = self._cell(inputs, state)
with tf.variable_scope(scope or self._name):
output = slim.layer_norm(output)
return output, res_state
elif self._apply_to == 'state':
output, res_state = self._cell(inputs, state)
with tf.variable_scope(scope or self._name):
res_state = slim.layer_norm(res_state)
return output, res_state
else:
raise ValueError('Unknown apply_to: "{}"'.format(self._apply_to))
# R-NEM CELL
示例4: _build_layer
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def _build_layer(inputs, layer):
# apply transformation
if layer['name'] == 'fc':
out = slim.fully_connected(inputs, layer['size'], activation_fn=None)
else:
raise KeyError('Unknown layer "{}"'.format(layer['name']))
# apply layer normalisation
if layer.get('ln', False):
out = slim.layer_norm(out)
# apply activation function
if layer.get('act', False):
out = ACTIVATION_FUNCTIONS[layer['act']](out)
return out
# NETWORK BUILDER
示例5: create_network_factory
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def create_network_factory(is_training, num_classes, add_logits,
weight_decay=1e-8, reuse=None):
def factory_fn(image):
with slim.arg_scope([slim.batch_norm, slim.dropout],
is_training=is_training):
with slim.arg_scope([slim.conv2d, slim.fully_connected,
slim.batch_norm, slim.layer_norm],
reuse=reuse):
features, logits = create_network(
image, num_classes=num_classes, add_logits=add_logits,
reuse=reuse, create_summaries=is_training,
weight_decay=weight_decay)
return features, logits
return factory_fn
示例6: _network_factory
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def _network_factory(weight_decay=1e-8):
def factory_fn(image, reuse):
with slim.arg_scope([slim.batch_norm, slim.dropout],
is_training=False):
with slim.arg_scope([slim.conv2d, slim.fully_connected,
slim.batch_norm, slim.layer_norm],
reuse=reuse):
features, logits = _create_network(
image, reuse=reuse, weight_decay=weight_decay)
return features, logits
return factory_fn
示例7: _call
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def _call(self, inputs, output_size, is_training):
inputs = self._subcall(inputs, output_size, is_training)
if self._spec.get('ln', False):
inputs = slim.layer_norm(inputs)
act = self._spec.get('act', False)
if act:
activation = ACTIVATION_FUNCTIONS[act]
return activation(inputs)
return inputs
示例8: resnet_arg_scope
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def resnet_arg_scope(weight_decay=0.0001,
activation_fn=tf.nn.relu,
use_layer_norm=True):
"""Defines the default ResNet arg scope.
TODO(gpapan): The batch-normalization related default values above are
appropriate for use in conjunction with the reference ResNet models
released at https://github.com/KaimingHe/deep-residual-networks. When
training ResNets from scratch, they might need to be tuned.
Args:
weight_decay: The weight decay to use for regularizing the model.
activation_fn: The activation function which is used in ResNet.
use_layer_norm: Whether or not to use layer normalization.
Returns:
An `arg_scope` to use for the resnet models.
"""
with slim.arg_scope(
[slim.conv2d],
weights_regularizer=slim.l2_regularizer(weight_decay),
weights_initializer=slim.variance_scaling_initializer(),
activation_fn=activation_fn,
normalizer_fn=slim.layer_norm if use_layer_norm else None,
normalizer_params=None):
# The following implies padding='SAME' for pool1, which makes feature
# alignment easier for dense prediction tasks. This is also used in
# https://github.com/facebook/fb.resnet.torch. However the accompanying
# code of 'Deep Residual Learning for Image Recognition' uses
# padding='VALID' for pool1. You can switch to that choice by setting
# slim.arg_scope([slim.max_pool2d], padding='VALID').
with slim.arg_scope([slim.max_pool2d], padding='SAME') as arg_sc:
return arg_sc
示例9: get_norm_layer
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def get_norm_layer(norm, training, updates_collections=None):
if norm == 'none':
return lambda x: x
elif norm == 'batch_norm':
return functools.partial(slim.batch_norm, scale=True, is_training=training, updates_collections=updates_collections)
elif norm == 'instance_norm':
return slim.instance_norm
elif norm == 'layer_norm':
return slim.layer_norm
示例10: bottleneck
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def bottleneck(inputs, depth, depth_bottleneck, stride, rate=1,
outputs_collections=None, scope=None):
"""Bottleneck residual unit variant with BN before convolutions.
This is the full preactivation residual unit variant proposed in [2]. See
Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck
variant which has an extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
Returns:
The ResNet unit's output.
"""
with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
preact = slim.layer_norm(inputs, activation_fn=tf.nn.relu, scope='preact')
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(preact, depth, [1, 1], stride=stride,
normalizer_fn=None, activation_fn=None,
scope='shortcut')
residual = slim.conv2d(preact, depth_bottleneck, [1, 1], stride=1,
scope='conv1')
residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,
rate=rate, scope='conv2')
residual = slim.conv2d(residual, depth, [1, 1], stride=1,
normalizer_fn=None, activation_fn=None,
scope='conv3')
output = shortcut + residual
return slim.utils.collect_named_outputs(outputs_collections,
sc.name,
output)
示例11: mpi_net
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def mpi_net(inputs, num_outputs, ngf=64, vscope='net', reuse_weights=False):
"""Network definition for multiplane image (MPI) inference.
Args:
inputs: stack of input images [batch, height, width, input_channels]
num_outputs: number of output channels
ngf: number of features for the first conv layer
vscope: variable scope
reuse_weights: whether to reuse weights (for weight sharing)
Returns:
pred: network output at the same spatial resolution as the inputs.
"""
with tf.variable_scope(vscope, reuse=reuse_weights):
with slim.arg_scope(
[slim.conv2d, slim.conv2d_transpose], normalizer_fn=slim.layer_norm):
cnv1_1 = slim.conv2d(inputs, ngf, [3, 3], scope='conv1_1', stride=1)
cnv1_2 = slim.conv2d(cnv1_1, ngf * 2, [3, 3], scope='conv1_2', stride=2)
cnv2_1 = slim.conv2d(cnv1_2, ngf * 2, [3, 3], scope='conv2_1', stride=1)
cnv2_2 = slim.conv2d(cnv2_1, ngf * 4, [3, 3], scope='conv2_2', stride=2)
cnv3_1 = slim.conv2d(cnv2_2, ngf * 4, [3, 3], scope='conv3_1', stride=1)
cnv3_2 = slim.conv2d(cnv3_1, ngf * 4, [3, 3], scope='conv3_2', stride=1)
cnv3_3 = slim.conv2d(cnv3_2, ngf * 8, [3, 3], scope='conv3_3', stride=2)
cnv4_1 = slim.conv2d(
cnv3_3, ngf * 8, [3, 3], scope='conv4_1', stride=1, rate=2)
cnv4_2 = slim.conv2d(
cnv4_1, ngf * 8, [3, 3], scope='conv4_2', stride=1, rate=2)
cnv4_3 = slim.conv2d(
cnv4_2, ngf * 8, [3, 3], scope='conv4_3', stride=1, rate=2)
# Adding skips
skip = tf.concat([cnv4_3, cnv3_3], axis=3)
cnv6_1 = slim.conv2d_transpose(
skip, ngf * 4, [4, 4], scope='conv6_1', stride=2)
cnv6_2 = slim.conv2d(cnv6_1, ngf * 4, [3, 3], scope='conv6_2', stride=1)
cnv6_3 = slim.conv2d(cnv6_2, ngf * 4, [3, 3], scope='conv6_3', stride=1)
skip = tf.concat([cnv6_3, cnv2_2], axis=3)
cnv7_1 = slim.conv2d_transpose(
skip, ngf * 2, [4, 4], scope='conv7_1', stride=2)
cnv7_2 = slim.conv2d(cnv7_1, ngf * 2, [3, 3], scope='conv7_2', stride=1)
skip = tf.concat([cnv7_2, cnv1_2], axis=3)
cnv8_1 = slim.conv2d_transpose(
skip, ngf, [4, 4], scope='conv8_1', stride=2)
cnv8_2 = slim.conv2d(cnv8_1, ngf, [3, 3], scope='conv8_2', stride=1)
feat = cnv8_2
pred = slim.conv2d(
feat,
num_outputs, [1, 1],
stride=1,
activation_fn=tf.nn.tanh,
normalizer_fn=None,
scope='color_pred')
return pred
示例12: build_network
# 需要导入模块: from tensorflow.contrib import slim [as 别名]
# 或者: from tensorflow.contrib.slim import layer_norm [as 别名]
def build_network(K, input, recurrent, output):
with tf.name_scope('inner_RNN'):
# build recurrent
for i, layer in enumerate(recurrent):
if layer['name'] == 'rnn':
cell = tf.contrib.rnn.BasicRNNCell(layer['size'], activation=ACTIVATION_FUNCTIONS['linear'])
cell = LayerNormWrapper(cell, apply_to='output', name='LayerNormR{}'.format(i)) if layer.get('ln') else cell
cell = ActivationFunctionWrapper(cell, activation=layer['act'], apply_to='state')
cell = ActivationFunctionWrapper(cell, activation=layer['act'], apply_to='output')
elif layer['name'] == 'lstm':
cell = tf.contrib.rnn.LayerNormBasicLSTMCell(layer['size'], layer_norm=layer.get('ln', False))
if layer.get('act'):
print("WARNING: activation function arg for LSTM Cell is ignored. Default (tanh) is used in stead.")
elif layer['name'] == 'r_nem':
cell = R_NEM(encoder=layer['encoder'],
core=layer['core'],
context=layer['context'],
attention=layer['attention'],
actions=layer.get('actions', None),
size=layer['size'],
K=K)
cell = LayerNormWrapper(cell, apply_to='output', name='LayerNormR{}'.format(i)) if layer.get('ln') else cell
cell = ActivationFunctionWrapper(cell, activation=layer['act'], apply_to='state')
cell = ActivationFunctionWrapper(cell, activation=layer['act'], apply_to='output')
else:
raise ValueError('Unknown recurrent name "{}"'.format(layer['name']))
# build input
for i, layer in reversed(list(enumerate(input))):
if layer['name'] == 'reshape':
cell = ReshapeWrapper(cell, layer['shape'], apply_to='input')
else:
cell = ActivationFunctionWrapper(cell, layer['act'], apply_to='input')
cell = LayerNormWrapper(cell, apply_to='input', name='LayerNormI{}'.format(i)) if layer.get('ln') else cell
cell = InputWrapper(cell, layer, name="InputWrapper{}".format(i))
# build output
for i, layer in enumerate(output):
if layer['name'] == 'reshape':
cell = ReshapeWrapper(cell, layer['shape'])
else:
n_out = layer.get('n_out', 1)
cell = OutputWrapper(cell, layer, n_out=n_out, name="OutputWrapper{}".format(i))
cell = LayerNormWrapper(cell, apply_to='output', name='LayerNormO{}'.format(i)) if layer.get('ln') else cell
cell = ActivationFunctionWrapper(cell, layer['act'], apply_to='output')
return cell