TensorFlow实现经典深度学习网络（2）：TensorFlow实现VGGNet

TensorFlow实现经典深度学习网络（2）：TensorFlow实现VGGNet


        VGGNet是经典的深度学习网络之一，由牛津大学的视觉几何组（Visual Geometry Group）提出，是ILSVRC-2014中定位任务第一名和分类任务第二名（top-5错误率7.3%,19层神经网络）。其探索了卷积神经网络的深度与其性能之间的关系，拓展性很强，通过反复堆叠3×3的小型卷积核和2×2的最大池化层，成功构筑了16～19层深的卷积神经网络，增加网络深度可以有效提升模型的效果，而且VGGNet对其他数据集具有很好的泛化能力。到目前为止，VGGNet依然常被用来提取图像特征。





        VGGNet论文中全部使用来3×3的卷积核和2×2的池化核，通过不断加深网络结构来提升性能。上图为VGGNet各级别的网络结构图和各级别网络参数量。VGGNet拥有5段卷积，每一段有2～3个卷积层，同时每段尾部连有最大池化层来缩小图片尺寸。VGGNet为了在公平的原则下探究网络深度对模型精确度的影响，所有卷积层有相同的配置，即卷积核大小为3x3，步长为1，填充为1；共有5个最大池化层，大小都为2x2，步长为2；共有三个全连接层，前两层都有4096通道，第三层共1000路及代表1000个标签类别；最后一层为softmax层；所有隐藏层后都带有ReLU非线性激活函数；经过实验证明，AlexNet中提出的局部响应归一化（LRN）对性能提升并没有什么帮助，而且还浪费了内存的计算的损耗。可以说，VGGNet使得CNN对特征的学习能力更强。

        VGGNet的突出影响包括：
       （1）一个大卷积核分解成连续多个小卷积核;

       （2）减少参数，降低计算，增加深度;

       （3）继承AlexNet结构特点：简单，有效;

       （4）网络改造的首选基础网络

        因使用ImageNet数据集非常耗时，因此本文会对完整的VGGNet网络进行速度测试，评测forward耗时和backward耗时。若读者感兴趣，可自行下载ImageNet数据集进行训练测试。

        在准备工作就绪后，我们就可以搭建VGGNet网络，也就是版本D，其他版本读者可仿照以下代码自行修改。以下代码是本人根据自己的理解整理而成，并亲自验证，代码中标有自己理解所得的注释，若有错误请指正。

# -*- coding: utf-8 -*-
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# VGGNet-16
# 导入常用库，载入TensorFlow
from datetime import datetime
import math
import time
import tensorflow as tf

# 创建卷积层并把本层的参数存入参数列表
def conv_op(input_op, name, kh, kw, n_out, dh, dw, p):
n_in = input_op.get_shape()[-1].value

with tf.name_scope(name) as scope:
kernel = tf.get_variable(scope + "w", shape=[kh, kw, n_in, n_out],
dtype=tf.float32,
initializer=tf.contrib.layers.xavier_initializer_conv2d())

# 对input_op进行卷积处理
conv = tf.nn.conv2d(input_op, kernel, (1, dh, dw, 1), padding='SAME')
bias_init_val = tf.constant(0.0, shape=[n_out], dtype=tf.float32)
biases = tf.Variable(bias_init_val, trainable=True, name='b')
z = tf.nn.bias_add(conv, biases)
activation = tf.nn.relu(z, name=scope)
p += [kernel, biases]
return activation

# 定义全连接层创建函数fc_op
def fc_op(input_op, name, n_out, p):
n_in = input_op.get_shape()[-1].value

with tf.name_scope(name) as scope:
kernel = tf.get_variable(scope + "w", shape=[n_in, n_out],
dtype=tf.float32,
initializer=tf.contrib.layers.xavier_initializer())
biases = tf.Variable(tf.constant(0.1, shape= [n_out],
dtype=tf.float32), name='b')
activation = tf.nn.relu_layer(input_op, kernel, biases, name=scope)
p += [kernel, biases]
return activation

# 定义最大池化层创建函数mpool_op
def mpool_op(input_op, name, kh, kw, dh, dw):
return tf.nn.max_pool(input_op,
ksize=[1, kh, kw, 1],
strides=[1, dh, dw, 1],
padding='SAME',
name=name)

# 创建VGGNet-16
def inference_op(input_op, keep_prob):

p = []

# conv1
conv1_1 = conv_op(input_op, name="conv1_1", kh=3, kw=3, n_out=64, dh=1, dw=1, p=p)
conv1_2 = conv_op(conv1_1, name="conv1_2", kh=3, kw=3, n_out=64, dh=1, dw=1, p=p)
pool1 = mpool_op(conv1_2, name="pool1", kh=2, kw=2, dw=2, dh=2)

# conv2
conv2_1 = conv_op(pool1, name="conv2_1", kh=3, kw=3, n_out=128, dh=1, dw=1, p=p)
conv2_2 = conv_op(conv2_1, name="conv2_2", kh=3, kw=3, n_out=128, dh=1, dw=1, p=p)
pool2 = mpool_op(conv2_2, name="pool2", kh=2, kw=2, dw=2, dh=2)

# conv3
conv3_1 = conv_op(pool2, name="conv3_1", kh=3, kw=3, n_out=256, dh=1, dw=1, p=p)
conv3_2 = conv_op(conv3_1, name="conv3_2", kh=3, kw=3, n_out=256, dh=1, dw=1, p=p)
conv3_3 = conv_op(conv3_2, name="conv3_3", kh=3, kw=3, n_out=256, dh=1, dw=1, p=p)
pool3 = mpool_op(conv3_3, name="pool3", kh=2, kw=2, dw=2, dh=2)

# conv4
conv4_1 = conv_op(pool3, name="conv4_1", kh=3, kw=3, n_out=512, dh=1, dw=1, p=p)
conv4_2 = conv_op(conv4_1, name="conv4_2", kh=3, kw=3, n_out=512, dh=1, dw=1, p=p)
conv4_3 = conv_op(conv4_2, name="conv4_3", kh=3, kw=3, n_out=2512, dh=1, dw=1, p=p)
pool4 = mpool_op(conv4_3, name="pool4", kh=2, kw=2, dw=2, dh=2)

# conv5
conv5_1 = conv_op(pool4, name="conv5_1", kh=3, kw=3, n_out=512, dh=1, dw=1, p=p)
conv5_2 = conv_op(conv5_1, name="conv5_2", kh=3, kw=3, n_out=512, dh=1, dw=1, p=p)
conv5_3 = conv_op(conv5_2, name="conv5_3", kh=3, kw=3, n_out=512, dh=1, dw=1, p=p)
pool5 = mpool_op(conv5_3, name="pool5", kh=2, kw=2, dw=2, dh=2)

# 将conv5输出结果扁平化
shp = pool5.get_shape()
flattened_shape = shp[1].value * shp[2].value * shp[3].value
resh1 = tf.reshape(pool5, [-1, flattened_shape], name="resh1")

# fc6
fc6 = fc_op(resh1, name="fc6", n_out=4096, p=p)
fc6_drop = tf.nn.dropout(fc6, keep_prob, name="fc6_drop")

# fc7
fc7 = fc_op(fc6_drop, name="fc7", n_out=4096, p=p)
fc7_drop = tf.nn.dropout(fc7, keep_prob, name="fc7_drop")

# fc8
fc8 = fc_op(fc7_drop, name="fc8", n_out=1000, p=p)
softmax = tf.nn.softmax(fc8)
predictions = tf.argmax(softmax, 1)
return predictions, softmax, fc8, p

# 评测
def time_tensorflow_run(session, target, feed, info_string):
num_steps_burn_in = 10
total_duration = 0.0
total_duration_squared = 0.0
for i in range(num_batches + num_steps_burn_in):
start_time = time.time()
_ = session.run(target, feed_dict=feed)
duration = time.time() - start_time
if not i % 10:
print('%s: step %d, duration = %.3f' % (datetime.now(), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / num_batches
vr = total_duration_squared / num_batches - mn *mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), info_string, num_batches, mn, sd))

# 定义评测主函数
def run_benchmark():
with tf.Graph().as_default():
image_size = 224
images = tf.Variable(tf.random_normal([batch_size,
image_size,
image_size, 3],
dtype=tf.float32,
stddev=1e-1))

keep_prob = tf.placeholder(tf.float32)
predictions, softmax, fc8, p = inference_op(images, keep_prob)

# 创建Session并初始化全局参数
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

time_tensorflow_run(sess, predictions, {keep_prob: 1.0}, "Forward")
objective = tf.nn.l2_loss(fc8)
grad = tf.gradients(objective, p)
time_tensorflow_run(sess, grad, {keep_prob: 0.5}, "Forward-backward")

batch_size = 32
num_batches = 100
run_benchmark()

        运行程序，我们会看到如下的程序显示（部分）

2017-10-13 15:26:22.435496: step -10, duration = 14.238
2017-10-13 15:28:45.836558: step 0, duration = 14.355
2017-10-13 15:31:08.099205: step 10, duration = 14.220
2017-10-13 15:33:30.337682: step 20, duration = 14.211
2017-10-13 15:35:52.644435: step 30, duration = 14.302
2017-10-13 15:38:15.698386: step 40, duration = 14.325
2017-10-13 15:40:38.672939: step 50, duration = 14.256
2017-10-13 15:43:01.125935: step 60, duration = 14.209
2017-10-13 15:45:23.317328: step 70, duration = 14.254
2017-10-13 15:47:45.705173: step 80, duration = 14.437
2017-10-13 15:50:13.805813: step 90, duration = 14.440
2017-10-13 15:52:26.331622: Forward across 100 steps, 1.572 +/- 4.473 sec / batch
2017-10-13 15:53:12.896681: step -10, duration = 46.497
2017-10-13 16:00:58.813022: step 0, duration = 46.705
2017-10-13 16:08:44.818871: step 10, duration = 46.518
2017-10-13 16:16:35.558625: step 20, duration = 46.913
2017-10-13 16:24:30.725427: step 30, duration = 46.850
2017-10-13 16:32:18.516636: step 40, duration = 46.793

        以上为程序运行过程中显示的VGGNet结构以及forward、backword运算时间。
[align=left]        至此，TensorFlow实现VGGNet的工作就完成了。VGG系列的卷积神经网络在ILSVRC2014比赛中最终达到来7.3%的错误率，相比AlexNet进步非常大，成为一种经典的卷积神经网络。后续，我将和大家探讨其他的经典深度学习网络。
[/align]

       在后续工作中，我将继续为大家展现TensorFlow和深度学习网络带来的无尽乐趣，我将和大家一起探讨深度学习的奥秘。当然，如果你感兴趣，我的Weibo将与你一起分享最前沿的人工智能、机器学习、深度学习与计算机视觉方面的技术。