脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - TensorFlow卷积神经网络AlexNet实现示例详解

TensorFlow卷积神经网络AlexNet实现示例详解

2022-02-24 00:15零尾 Python

这篇文章主要为大家介绍了TensorFlow卷积神经网络AlexNet实现示例详解,有需要的朋友可以借鉴参考下,希望能够有所帮助,祝大家多多进步

2012年,Hinton的学生Alex Krizhevsky提出了深度卷积神经网络模型AlexNet,它可以算是LeNet的一种更深更宽的版本。AlexNet以显著的优势赢得了竞争激烈的ILSVRC 2012比赛,top-5的错误率降低至了16.4%,远远领先第二名的26.2%的成绩。AlexNet的出现意义非常重大,它证明了CNN在复杂模型下的有效性,而且使用GPU使得训练在可接受的时间范围内得到结果,让CNN和GPU都大火了一把。AlexNet可以说是神经网络在低谷期后的第一次发声,确立了深度学习(深度卷积网络)在计算机视觉的统治地位,同时也推动了深度学习在语音识别、自然语言处理、强化学习等领域的拓展。

模型结构

TensorFlow卷积神经网络AlexNet实现示例详解

整个AlexNet前5层为卷积层,后3个为全连接层,其中最后一层是1000类输出的Softmax分类输出层。LRN层在第1个及第2个卷积层之后,Max pooling层在两个LRN层之后和最后一个卷积层之后。ReLU激活函数跟在5个卷积层和2个全连接层后面(最后输出层没有)。

因为AlexNet训练时使用了两块GPU,因此这个结构图中不少组件都被拆为了两部分。现在我们GPU的显存足够大可以放下全部模型参数,因此只考虑一块GPU的情况。

AlexNet中包含的新技术点

成功使用ReLU作为CNN的激活函数,并验证了在较深的网络中其效果超过了Sigmoid,成功解决了Sigmoid在网络较深时的梯度消失问题。虽然ReLU在很久之前就被提出来了,但AlexNet的出现才将其发扬光大。

网络中加入了Dropout层,训练时使用Dropout随机杀死(忽略)一部分神经元,以避免模型过拟合。(AlexNet通过实践证实了Dropout的效果,关于Dropout有单独的论文论述。)

池化层使用Max pooling,此前CNN普遍使用平均池化,最大池化避免了平均池化的模糊化效果。且提出让步长小于池化核的尺寸,这样池化层的输出之间会有重叠和覆盖,提升了特征的丰富性。

提出了LRN层(Local Response Normalization),对局部神经元的活动创建竞争机制,使得其中响应比较大的值变的相对更大,并抑制其它反馈较小的神经元,增强了模型的泛化能力。

使用CUDA加速深度神经网络的训练,利用GPU强大的并行计算能力,处理神经网络训练时大量的矩阵运算。(当时Alex使用的是两块GTX 580,单个GPU只有3GB显存,限制了可训练的网络的最大规模。因此他将AlexNet分布在两个GPU上,每个GPU的显存中储存一半的神经元的参数。现在GTX 1080Ti都出来了,硬件发展的还是比较快的,这也是深度学习能飞速发展的重要原因之一吧)

数据增强。随机地从256*256的原始图像中截取224*224大小的区域(以及水平翻转的镜像),相当于增加了((256-224)^2)*2=2048倍的数据量。如果没有数据增强,仅靠原始数据的数据量,参数众多的CNN会陷入过拟合中,进行数据增强后可以大大减轻过拟合,提升模型的泛化能力。进行预测时,则是取图片的四个角加中间共5个位置,并进行左右翻转,一共获得10个图片,对它们进行预测并对求10次结果求均值。同时,AlexNet论文中提到了会对图像的RGB数据进行PCA处理,并对主成分做一个标准差为0.1的高斯扰动,增加了一些噪声,这个Trick可以让错误率再下降1%。

AlexNet每层的超参数

其中,

Input:图片尺寸224*224

Conv1:卷积核11*11,步长4,96个filter(卷积核尺寸较大)

ReLU

LRN1

Max pooling1:3*3,步长2

Conv2:卷积核5*5,步长1,256个filter

ReLU

LRN2

Max pooling2:3*3,步长2

Conv3:卷积核3*3,步长1,384个filter

ReLU

Conv4:卷积核3*3,步长1,384个filter

ReLU

Conv5:卷积核3*3,步长1,256个filter

ReLU

Max pooling3:3*3,步长2

FC1:4096

ReLU

FC2:4096

ReLU

FC3(Output):1000

对AlexNet使用MNIST数据集

代码如下:

?
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
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
from datetime import datetime
import time
import tensorflow as tf
import input_data
mnist = input_data.read_data_sets('data/', one_hot=True)
print("MNIST READY")
# 定义网络参数
n_input = 784 # 输入的维度
n_output = 10 # 标签的维度
learning_rate = 0.001
dropout = 0.75
# 定义函数print_activations来显示网络每一层结构,展示每一个卷积层或池化层输出tensor的尺寸
def print_activations(t):
    print(t.op.name, '', t.get_shape().as_list())
# 定义卷积操作
def conv2d(input, w, b):
    return tf.nn.relu(tf.nn.bias_add(tf.nn.conv2d(input, w, strides=[1, 1, 1, 1], padding='SAME'), b)) # 参数分别指定了卷积核的尺寸、多少个channel、filter的个数即产生特征图的个数                                                                                       # 步长为1,即扫描全图像素,[1, 1, 1, 1]分别代表batch_size、h、w、c的stride
# 定义池化操作
def max_pool(input):
    return tf.nn.max_pool(input, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') # padding有两种选择:'SAME'(窗口滑动时,像素不够会自动补0)或'VALID'(不够就跳过)两种选择
# 定义全连接操作
def fc(input, w, b):
    return tf.nn.relu(tf.add(tf.matmul(input, w), b)) # w*x+b,再通过非线性激活函数relu
# 定义网络结构
def alex_net(_input, _weights, _biases, _keep_prob):
    _input_r = tf.reshape(_input, [-1, 28, 28, 1])  # 对图像做一个预处理,转换为tf支持的格式,即[n, h, w, c],-1是确定好其它3维后,让tf去推断剩下的1维
    with tf.name_scope('conv1'):
        _conv1 = conv2d(_input_r, _weights['wc1'], _biases['bc1'])
        print_activations(_conv1) # 将这一层最后输出的tensor conv1的结构打印出来
 
    # # 这里参数基本都是AlexNet论文中的推荐值,但目前其他经典卷积神经网络模型基本都放弃了LRN(主要是效果不明显),
    # # 并且使用LRN也会让前馈、反馈的速度大大下降(整体速度降到1/3)
    # with tf.name_scope('_lrn1'):
    #     _lrn1 = tf.nn.lrn(_conv1, 4, bias=1.0, alpha=0.001/9, beta=0.75)
 
    with tf.name_scope('pool1'):
        _pool1 = max_pool(_conv1)
        print_activations(_pool1)
 
    with tf.name_scope('conv2'):
        _conv2 = conv2d(_pool1, _weights['wc2'], _biases['bc2'])
        print_activations(_conv2)
 
    # with tf.name_scope('_lrn2'):
    #     _lrn2 = tf.nn.lrn(_conv2, 4, bias=1.0, alpha=0.001/9, beta=0.75)
 
    with tf.name_scope('pool2'):
        _pool2 = max_pool(_conv2)
        print_activations(_pool2)
 
    with tf.name_scope('conv3'):
        _conv3 = conv2d(_pool2, _weights['wc3'], _biases['bc3'])
        print_activations(_conv3)
 
    with tf.name_scope('conv4'):
        _conv4 = conv2d(_conv3, _weights['wc4'], _biases['bc4'])
        print_activations(_conv4)
 
    with tf.name_scope('conv5'):
        _conv5 = conv2d(_conv4, _weights['wc5'], _biases['bc5'])
        print_activations(_conv5)
 
    with tf.name_scope('pool3'):
        _pool3 = max_pool(_conv5)
        print_activations(_pool3)
 
    _densel = tf.reshape(_pool3, [-1, _weights['wd1'].get_shape().as_list()[0]])  # 定义全连接层的输入,把pool2的输出做一个reshape,变为向量的形式
 
    # pool_shape = _pool3.get_shape().as_list()
    # nodes = pool_shape[1] * pool_shape[2] * pool_shape[3]
 
    with tf.name_scope('fc1'):
        _fc1 = fc(_densel, _weights['wd1'], _biases['bd1'])
        _fc1_drop = tf.nn.dropout(_fc1, _keep_prob) # 为了减轻过拟合,使用Dropout层
        print_activations(_fc1_drop)
 
    with tf.name_scope('fc2'):
        _fc2 = fc(_fc1_drop, _weights['wd2'], _biases['bd2'])
        _fc2_drop = tf.nn.dropout(_fc2, _keep_prob)
        print_activations(_fc2_drop)
 
    with tf.name_scope('out'):
        _out = tf.add(tf.matmul(_fc2_drop, _weights['wd3']), _biases['bd3'])
        print_activations(_out)
 
    return _out
print("CNN READY")
 
x = tf.placeholder(tf.float32, [None, n_input]) # 用placeholder先占地方,样本个数不确定为None
y = tf.placeholder(tf.float32, [None, n_output]) # 用placeholder先占地方,样本个数不确定为None
keep_prob = tf.placeholder(tf.float32)
 
# 存储所有的网络参数
weights = {
    # 使用截断的正态分布(标准差0.1)初始化卷积核的参数kernel,卷积核大小为3*3,channel为1,个数64
    'wc1': tf.Variable(tf.truncated_normal([3, 3, 1, 64], dtype=tf.float32, stddev=0.1), name='weights1'),
    'wc2': tf.Variable(tf.truncated_normal([3, 3, 64, 128], dtype=tf.float32, stddev=0.1), name='weights2'),
    'wc3': tf.Variable(tf.truncated_normal([3, 3, 128, 256], dtype=tf.float32, stddev=0.1), name='weights3'),
    'wc4': tf.Variable(tf.truncated_normal([3, 3, 256, 256], dtype=tf.float32, stddev=0.1), name='weights4'),
    'wc5': tf.Variable(tf.truncated_normal([3, 3, 256, 128], dtype=tf.float32, stddev=0.1), name='weights5'),
    'wd1': tf.Variable(tf.truncated_normal([4*4*128, 1024], dtype=tf.float32, stddev=0.1), name='weights_fc1'),
    'wd2': tf.Variable(tf.random_normal([1024, 1024], dtype=tf.float32, stddev=0.1), name='weights_fc2'),
    'wd3': tf.Variable(tf.random_normal([1024, n_output], dtype=tf.float32, stddev=0.1), name='weights_output')
}
biases = {
    'bc1': tf.Variable(tf.constant(0.0, shape=[64], dtype=tf.float32), trainable=True, name='biases1'),
    'bc2': tf.Variable(tf.constant(0.0, shape=[128], dtype=tf.float32), trainable=True, name='biases2'),
    'bc3': tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32), trainable=True, name='biases3'),
    'bc4': tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32), trainable=True, name='biases4'),
    'bc5': tf.Variable(tf.constant(0.0, shape=[128], dtype=tf.float32), trainable=True, name='biases5'),
    'bd1': tf.Variable(tf.constant(0.0, shape=[1024], dtype=tf.float32), trainable=True, name='biases_fc1'),
    'bd2': tf.Variable(tf.constant(0.0, shape=[1024], dtype=tf.float32), trainable=True, name='biases_fc2'),
    'bd3': tf.Variable(tf.constant(0.0, shape=[n_output], dtype=tf.float32), trainable=True, name='biases_output')
}
 
pred = alex_net(x, weights, biases, keep_prob) # 前向传播的预测值
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(pred, y)) # 交叉熵损失函数,参数分别为预测值_pred和实际label值y,reduce_mean为求平均loss
optm = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost) # 梯度下降优化器
corr = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1)) # tf.equal()对比预测值的索引和实际label的索引是否一样,一样返回True,不一样返回False
accuracy = tf.reduce_mean(tf.cast(corr, tf.float32)) # 将pred即True或False转换为1或0,并对所有的判断结果求均值
# 初始化所有参数
init = tf.global_variables_initializer()
print("FUNCTIONS READY")
 
# 上面神经网络结构定义好之后,下面定义一些超参数
training_epochs = 1000 # 所有样本迭代1000次
batch_size = 1 # 每进行一次迭代选择50个样本
display_step = 10
 
sess = tf.Session() # 定义一个Session
sess.run(init) # 在sess里run一下初始化操作
for epoch in range(training_epochs):
    avg_cost = 0.
    total_batch = int(mnist.train.num_examples/batch_size)
    start_time = time.time()
    for i in range(total_batch):
        batch_xs, batch_ys = mnist.train.next_batch(batch_size) # 逐个batch的去取数据
        # 获取批数据
        sess.run(optm, feed_dict={x: batch_xs, y: batch_ys, keep_prob:dropout})
        avg_cost += sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keep_prob:1.0})/total_batch
    if epoch % display_step == 0:
        train_accuracy = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys, keep_prob: 1.0})
        test_accuracy = sess.run(accuracy, feed_dict={x: mnist.test.images, y: mnist.test.labels, keep_prob:1.0})
        print("Epoch: %03d/%03d cost: %.9f TRAIN ACCURACY: %.3f TEST ACCURACY: %.3f" % (epoch, training_epochs, avg_cost, train_accuracy, test_accuracy))
    # 计算每轮迭代的平均耗时mn和标准差sd,并显示
    duration = time.time() - start_time
    print('%s: step %d, duration = %.3f' % (datetime.now(), epoch, duration))
 
print("DONE")

本文参考《TensorFlow实战》黄文坚 唐源 著

以上就是TensorFlow卷积神经网络AlexNet实现示例详解的详细内容,更多关于TensorFlow卷积神经网络AlexNet的资料请关注服务器之家其它相关文章!

原文链接:https://blog.csdn.net/lwplwf/article/details/72870378

延伸 · 阅读

精彩推荐