Spaces:
Sleeping
Sleeping
File size: 3,863 Bytes
5672777 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 |
# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Neural network operations commonly shared by the architectures."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import functools
import tensorflow as tf, tf_keras
class NormActivation(tf_keras.layers.Layer):
"""Combined Normalization and Activation layers."""
def __init__(self,
momentum=0.997,
epsilon=1e-4,
trainable=True,
init_zero=False,
use_activation=True,
activation='relu',
fused=True,
name=None):
"""A class to construct layers for a batch normalization followed by a ReLU.
Args:
momentum: momentum for the moving average.
epsilon: small float added to variance to avoid dividing by zero.
trainable: `bool`, if True also add variables to the graph collection
GraphKeys.TRAINABLE_VARIABLES. If False, freeze batch normalization
layer.
init_zero: `bool` if True, initializes scale parameter of batch
normalization with 0. If False, initialize it with 1.
use_activation: `bool`, whether to add the optional activation layer after
the batch normalization layer.
activation: 'string', the type of the activation layer. Currently support
`relu` and `swish`.
fused: `bool` fused option in batch normalziation.
name: `str` name for the operation.
"""
super(NormActivation, self).__init__(trainable=trainable)
if init_zero:
gamma_initializer = tf_keras.initializers.Zeros()
else:
gamma_initializer = tf_keras.initializers.Ones()
self._normalization_op = tf_keras.layers.BatchNormalization(
momentum=momentum,
epsilon=epsilon,
center=True,
scale=True,
trainable=trainable,
fused=fused,
gamma_initializer=gamma_initializer,
name=name)
self._use_activation = use_activation
if activation == 'relu':
self._activation_op = tf.nn.relu
elif activation == 'swish':
self._activation_op = tf.nn.swish
else:
raise ValueError('Unsupported activation `{}`.'.format(activation))
def __call__(self, inputs, is_training=None):
"""Builds the normalization layer followed by an optional activation layer.
Args:
inputs: `Tensor` of shape `[batch, channels, ...]`.
is_training: `boolean`, if True if model is in training mode.
Returns:
A normalized `Tensor` with the same `data_format`.
"""
# We will need to keep training=None by default, so that it can be inherit
# from keras.Model.training
if is_training and self.trainable:
is_training = True
inputs = self._normalization_op(inputs, training=is_training)
if self._use_activation:
inputs = self._activation_op(inputs)
return inputs
def norm_activation_builder(momentum=0.997,
epsilon=1e-4,
trainable=True,
activation='relu',
**kwargs):
return functools.partial(
NormActivation,
momentum=momentum,
epsilon=epsilon,
trainable=trainable,
activation=activation,
**kwargs)
|