一、网络结构
- conv1:卷积层和ReLU激活函数
- pool1:最大池化
- norm1:LRN
- conv2:卷积层和ReLU激活函数
- norm2:LRN
- pool2:最大池化
- local3:全连接层和ReLU激活函数
- local4:全连接层和ReLU激活函数
- logits:模型Inference的输出结果
1,2,3是第一个卷积层。4,5,6是第二个卷积层。7是第一个全连接层,8是第二个全连接层,9是输出层。
二、TensorFlow实现进阶的卷积网络
我们使用的数据集是CIFAR-10,包含60000张的彩色图像,其中训练集50000张,测试集10000张。此数据集一共标注10类,分别是airplane、automobile、bird、cat、deer、dog、frog、horse、ship和truck。
首先,下载TensorFlow Models库,使用其中提供的CIFAR-10数据的类。
git clone https://github.com/tensorflow/models.git
cd models/tutorials/image/cifar10
然后,开始我们的程序。
import cifar10, cifar10_input
import tensorflow as tf
import numpy as np
import time
# 定义batch_size, 训练轮数, 以及数据存储路径
max_steps = 3000
batch_size = 128
data_dir = '/tmp/cifar10_data/cifar-10-batches-bin'
# 初始化权重,并增加了L2正则化
def variable_with_weight_loss(shape, stddev, w1):
var = tf.Variable(tf.truncated_normal(shape, stddev=stddev))
if w1 is not None:
weight_loss = tf.multiply(tf.nn.l2_loss(var), w1, name='weight_loss')
tf.add_to_collection('losses', weight_loss)
return var
# 下载数据,并产生训练数据和测试数据
tf.app.flags.DEFINE_string('f', '', 'kernel')
cifar10.maybe_download_and_extract()
images_train, labels_train = cifar10_input.distorted_inputs(data_dir=data_dir, batch_size=batch_size) # 产生训练数据
images_test, labels_test = cifar10_input.inputs(eval_data=True, data_dir=data_dir, batch_size=batch_size) # 产生测试数据
# 创建输入数据和label的占位符
image_holder = tf.placeholder(tf.float32, [batch_size, 24, 24, 3])
label_holder = tf.placeholder(tf.int32, [batch_size])
# 第一个卷积层
weight1 = variable_with_weight_loss(shape=[5, 5, 3, 64], stddev=5e-2, w1=0.0) # 权重
kernel1 = tf.nn.conv2d(image_holder, weight1, [1, 1, 1, 1], padding="SAME") # 卷积
bias1 = tf.Variable(tf.constant(0.0, shape=[64])) # 偏置
conv1 = tf.nn.relu(tf.nn.bias_add(kernel1, bias1)) # 激活
pool1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 2, 2,1], padding='SAME') # 池化
norm1 = tf.nn.lrn(pool1, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75) # 局部响应归一化
# 第二个卷积层
weight2 = variable_with_weight_loss(shape=[5, 5, 64, 64], stddev=5e-2, w1=0.0) # 权重
kernel2 = tf.nn.conv2d(norm1, weight2, [1, 1, 1, 1], padding="SAME") # 卷积
bias2 = tf.Variable(tf.constant(0.1, shape=[64]))
conv2 = tf.nn.relu(tf.nn.bias_add(kernel2, bias2))
norm2 = tf.nn.lrn(conv2, 4, bias=1.0, alpha=0.001 / 9.0, beta=0.75)
pool2 = tf.nn.max_pool(norm2, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME')
# 重设1D
reshape = tf.reshape(pool2, [batch_size, -1])
# 获取维度
dim = reshape.get_shape()[1].value
# 第三层:全连接层
weight3 = variable_with_weight_loss(shape=[dim, 384], stddev=0.04, w1=0.004)
bias3 = tf.Variable(tf.constant(0.1, shape=[384]))
local3 = tf.nn.relu(tf.matmul(reshape, weight3) + bias3)
# 第四层:全连接层
weight4 = variable_with_weight_loss(shape=[384, 192], stddev=0.04, w1=0.004)
bias4 = tf.Variable(tf.constant(0.1, shape=[192]))
local4 = tf.nn.relu(tf.matmul(local3, weight4) + bias4)
# 第五层:输出层
weight5 = variable_with_weight_loss(shape=[192, 10], stddev=1 / 192.0, w1=0.0)
bias5 = tf.Variable(tf.constant(0.0, shape=[10]))
logits = tf.add(tf.matmul(local4, weight5), bias5)
# 损失
def loss(logits, labels):
labels = tf.cast(labels, tf.int64)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='corss_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
return tf.add_n(tf.get_collection('losses'), name='total_loss')
# 获取损失值
loss = loss(logits, label_holder)
train_op = tf.train.AdamOptimizer(1e-3).minimize(loss)
# 取出分类最高的那类准确率
top_k_op = tf.nn.in_top_k(logits, label_holder, 1)
# 创建session
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
# 启动线程加速
tf.train.start_queue_runners()
# 开始训练
for step in range(max_steps):
start_time = time.time()
# 获取图像数据
image_batch, label_batch = sess.run([images_train, labels_train])
# 开始训练,计算损失
_, loss_value = sess.run([train_op, loss], feed_dict={image_holder: image_batch, label_holder: label_batch})
duration = time.time() - start_time
if step % 10 == 0:
examples_per_sec = batch_size / duration
sec_per_batch = float(duration)
format_str = ('step %d, loss=%.2f (%.1f examples/sec; %.3f sec/batch)')
print(format_str % (step, loss_value, examples_per_sec, sec_per_batch))
# 测试数据集上
num_examples = 10000
import math
num_iter = int(math.ceil(num_examples / batch_size))
true_count = 0
total_sample_count = num_iter * batch_size
step = 0
while step < num_iter:
image_batch, label_batch = sess.run([images_test, labels_test])
predictions = sess.run([top_k_op], feed_dict={image_holder: image_batch, label_holder: label_batch})
true_count += np.sum(predictions)
step += 1
precision = true_count / total_sample_count
print('precision @ 1 = %.3f' % precision)
最后,在测试集上取得了72.7%的错误率。