"""Tensorflow implementation of Face-ResNet:
A. Hasnat, J. Bohne, J. Milgram, S. Gentric, and L. Chen. Deepvisage: Making face
recognition simple yet with powerful generalization skills. arXiv:1703.08388, 2017.
"""
# MIT License
#
# Copyright (c) 2018 Yichun Shi
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import tensorflow as tf
import tensorflow.contrib.slim as slim
def parametric_relu(x):
num_channels = x.shape[-1].value
with tf.variable_scope('p_re_lu'):
alpha = tf.get_variable('alpha', (1,1,num_channels),
initializer=tf.constant_initializer(0.0),
dtype=tf.float32)
return tf.nn.relu(x) + alpha * tf.minimum(0.0, x)
# activation = lambda x: tf.keras.layers.PReLU(shared_axes=[1,2]).apply(x)
activation = parametric_relu
def se_module(input_net, ratio=16, reuse = None, scope = None):
with tf.variable_scope(scope, 'SE', [input_net], reuse=reuse):
h,w,c = tuple([dim.value for dim in input_net.shape[1:4]])
assert c % ratio == 0
hidden_units = int(c / ratio)
squeeze = slim.avg_pool2d(input_net, [h,w], padding='VALID')
excitation = slim.flatten(squeeze)
excitation = slim.fully_connected(excitation, hidden_units, scope='se_fc1',
weights_regularizer=None,
# weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
weights_initializer=slim.xavier_initializer(),
activation_fn=tf.nn.relu)
excitation = slim.fully_connected(excitation, c, scope='se_fc2',
weights_regularizer=None,
# weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
weights_initializer=slim.xavier_initializer(),
activation_fn=tf.nn.sigmoid)
excitation = tf.reshape(excitation, [-1,1,1,c])
output_net = input_net * excitation
return output_net
def conv_module(net, num_res_layers, num_kernels, reuse = None, scope = None):
with tf.variable_scope(scope, 'conv', [net], reuse=reuse):
# Every 2 conv layers constitute a residual block
if scope == 'conv1':
for i in range(len(num_kernels)):
with tf.variable_scope('layer_%d'%i, reuse=reuse):
net = slim.conv2d(net, num_kernels[i], kernel_size=3, stride=1, padding='VALID',
weights_initializer=slim.xavier_initializer())
print('| ---- layer_%d' % i)
net = slim.max_pool2d(net, 2, stride=2, padding='VALID')
else:
shortcut = net
for i in range(num_res_layers):
with tf.variable_scope('layer_%d'%i, reuse=reuse):
net = slim.conv2d(net, num_kernels[0], kernel_size=3, stride=1, padding='SAME',
weights_initializer=tf.truncated_normal_initializer(stddev=0.01),
biases_initializer=None)
print('| ---- layer_%d' % i)
if i % 2 == 1:
net = se_module(net)
net = net + shortcut
shortcut = net
print('| shortcut')
# Pooling for conv2 - conv4
if len(num_kernels) > 1:
with tf.variable_scope('expand', reuse=reuse):
net = slim.conv2d(net, num_kernels[1], kernel_size=3, stride=1, padding='VALID',
weights_initializer=slim.xavier_initializer())
net = slim.max_pool2d(net, 2, stride=2, padding='VALID')
print('- expand')
return net
def inference(images, keep_probability, phase_train=True, bottleneck_layer_size=512,
weight_decay=0.0, reuse=None, model_version=None):
with slim.arg_scope([slim.conv2d, slim.fully_connected],
weights_regularizer=slim.l2_regularizer(weight_decay),
activation_fn=activation,
normalizer_fn=None,
normalizer_params=None):
with tf.variable_scope('FaceResNet', [images], reuse=reuse):
with slim.arg_scope([slim.batch_norm, slim.dropout],
is_training=phase_train):
print('input shape:', [dim.value for dim in images.shape])
net = conv_module(images, 0, [32, 64], scope='conv1')
print('module_1 shape:', [dim.value for dim in net.shape])
net = conv_module(net, 2, [64, 128], scope='conv2')
print('module_2 shape:', [dim.value for dim in net.shape])
net = conv_module(net, 4, [128, 256], scope='conv3')
print('module_3 shape:', [dim.value for dim in net.shape])
net = conv_module(net, 10, [256, 512], scope='conv4')
print('module_4 shape:', [dim.value for dim in net.shape])
net = conv_module(net, 6, [512], scope='conv5')
print('module_5 shape:', [dim.value for dim in net.shape])
net = slim.flatten(net)
net = slim.fully_connected(net, bottleneck_layer_size, scope='Bottleneck',
weights_initializer=slim.xavier_initializer(),
activation_fn=None)
return net
