在之前的tf学习笔记系列中已经吧AlexNet的实现做了记录，这里继续，实现一下VGG。
参考之前分享过的论文详解log：深度学习VGG模型核心拆解
不过在自己实现之前，先来看看slim里的实现是什么样子的。
def vgg_16(inputs,
num_classes=1000,
is_training=True,
dropout_keep_prob=0.5,
spatial_squeeze=True,
scope='vgg_16',
fc_conv_padding='VALID'):
"""Oxford Net VGG 16-Layers version D Example.
Note: All the fully_connected layers have been transformed to conv2d layers.
To use in classification mode, resize input to 224x224.
Args:
inputs: a tensor of size [batch_size, height, width, channels].
num_classes: number of predicted classes.
is_training: whether or not the model is being trained.
dropout_keep_prob: the probability that activations are kept in the dropout
layers during training.
spatial_squeeze: whether or not should squeeze the spatial dimensions of the
outputs. Useful to remove unnecessary dimensions for classification.
scope: Optional scope for the variables.
fc_conv_padding: the type of padding to use for the fully connected layer
that is implemented as a convolutional layer. Use 'SAME' padding if you
are applying the network in a fully convolutional manner and want to
get a prediction map downsampled by a factor of 32 as an output. Otherwise,
the output prediction map will be (input / 32) - 6 in case of 'VALID' padding.
Returns:
the last op containing the log predictions and end_points dict.
"""
with tf.variable_scope(scope, 'vgg_16', [inputs]) as sc:
end_points_collection = sc.name + '_end_points'
# Collect outputs for conv2d, fully_connected and max_pool2d.
with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],
outputs_collections=end_points_collection):
net = slim.repeat(inputs, 2, slim.conv2d, 64, [3, 3], scope='conv1')
net = slim.max_pool2d(net, [2, 2], scope='pool1')
net = slim.repeat(net, 2, slim.conv2d, 128, [3, 3], scope='conv2')
net = slim.max_pool2d(net, [2, 2], scope='pool2')
net = slim.repeat(net, 3, slim.conv2d, 256, [3, 3], scope='conv3')
net = slim.max_pool2d(net, [2, 2], scope='pool3')
net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv4')
net = slim.max_pool2d(net, [2, 2], scope='pool4')
net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv5')
net = slim.max_pool2d(net, [2, 2], scope='pool5')
# Use conv2d instead of fully_connected layers.
net = slim.conv2d(net, 4096, [7, 7], padding=fc_conv_padding, scope='fc6')
net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
scope='dropout6')
net = slim.conv2d(net, 4096, [1, 1], scope='fc7')
net = slim.dropout(net, dropout_keep_prob, is_training=is_training,
scope='dropout7')
net = slim.conv2d(net, num_classes, [1, 1],
activation_fn=None,
normalizer_fn=None,
scope='fc8')
# Convert end_points_collection into a end_point dict.
end_points = slim.utils.convert_collection_to_dict(end_points_collection)
if spatial_squeeze:
net = tf.squeeze(net, [1, 2], name='fc8/squeezed')
end_points[sc.name + '/fc8'] = net
return net, end_points
怎么样，很简洁的吧。对的，由于slim封装了很多基础操作，可以精简代码，而且，slim.repeat()的API让创建多个相同layer变得更容易。
如果是我们自己封装实现的话，可以借鉴slim这种方式。
这里沿用之前AlexNet采用的方式来实现一下VGG16的基础版本。对比一下slim里最后三层都用卷积层替代了。
import tensorflow as tf
class VGG16(object):
'''
#use like this:
model = VGG16(input, num_classes, keep_prob, is_training)
score = model.fc8
# then you can get loss op using score
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits = score, labels = y))
'''
def __init__(self, input, num_classes, keep_prob=0.5, is_training=True):
self.INPUT = input
self.NUM_CLASSES = num_classes
self.KEEP_PROB = keep_prob
self.IS_TRAINING = is_training
self.default_image_size = 224
self.create()
def create(self):
with tf.name_scope('vgg16'):
x = conv('conv1_1', self.INPUT, 64)
x = conv('conv1_2', x, 64)
x = max_pool('pool1', x)
x = conv('conv2_1', x, 128)
x = conv('conv2_2', x, 128)
x = max_pool('pool2', x)
x = conv('conv3_1', x, 256)
x = conv('conv3_2', x, 256)
x = conv('conv3_3', x, 256)
x = max_pool('pool3', x)
x = conv('conv4_1', x, 512)
x = conv('conv4_2', x, 512)
x = conv('conv4_3', x, 512)
x = max_pool('pool4', x)
x = conv('conv5_1', x, 512)
x = conv('conv5_2', x, 512)
x = conv('conv5_3', x, 512)
x = max_pool('pool5', x)
x = fc('fc6', x, 4096)
x = dropout(x, self.KEEP_PROB)
x = fc('fc7', x, 4096)
x = dropout(x, self.KEEP_PROB)
self.fc8 = fc('fc8', x, self.NUM_CLASSES, is_relu=False)
def conv(name, input, num_filters, filter_height=3, filter_width=3, stride_x=1, stride_y=1, padding='SAME' ):
'''
先定义conv的通用模式
'''
input_channels = int(input.get_shape()[-1])
with tf.variable_scope(name) as scope:
weights = tf.get_variable('weights', shape=[filter_height, filter_width, input_channels, num_filters])
biases = tf.get_variable('biases', shape=[num_filters])
conv = tf.nn.conv2d(input, weights,
strides=[1, stride_y, stride_x, 1],
padding=padding)
bias = tf.nn.bias_add(conv, biases)
relu = tf.nn.relu(bias, name=scope.name)
return relu
def fc(name, input, num_output, is_relu=True, is_trainable=True):
'''定义全连接层'''
shape = input.get_shape()
if len(shape) == 4:
num_input = shape[1].value * shape[2].value * shape[3].value
else:
num_input = shape[-1].value
with tf.variable_scope(name) as scope:
weights = tf.get_variable('weights', shape=[num_input, num_output], trainable=is_trainable)
biases = tf.get_variable('biases', [num_output], trainable=is_trainable)
flat_x = tf.reshape(input, [-1, num_input])
act = tf.nn.xw_plus_b(flat_x, weights, biases, name=scope.name)
if is_relu:
return tf.nn.relu(act, name=scope.name)
else:
return act
def max_pool(name, input, filter_height=2, filter_width=2, stride_x=2, stride_y=2, padding='SAME'):
return tf.nn.max_pool(input,
ksize=[1, filter_height, filter_width, 1],
strides=[1, stride_x, stride_y, 1],
padding=padding,
name=name)
def dropout(input, keep_prob=0.5, is_training=True):
if is_training:
return tf.nn.dropout(input, keep_prob)
else:
return input
网友评论