TensorFlow 实现CNN自编码网络
1. 实现：
实现卷积编码和反卷积解码的自编码网络
2. 数据集：
MNIST数据集
3. 实现流程：
3.1 编码：CNN卷积操作
| CNN卷积操作 |
输入图像shape |
**函数 |
卷积核shape |
步长 |
输出图像shape |
| 第一层卷积 |
[batch,28,28,1] |
LeakReLU |
[3,3,1,16] |
[1,2,2,1] |
[batch,14,14,16] |
| 第二层卷积 |
[batch,14,14,16] |
LeakReLU |
[3,316,32] |
[1,2,2,1] |
[batch,7,7,32] |
| 第三层卷积 |
[batch,7,7,32] |
LeakReLU |
[3,3,32,64] |
[1,2,2,1] |
[batch,4,4,64] |
| 第四层卷积 |
[batch,4,4,64] |
LeakReLU |
[2,2,64,64] |
[1,2,2,1] |
[batch,2,2,64] |
3.2 解码：反卷积操作
| CNN反卷积操作 |
输入图像shape |
**函数 |
卷积核shape |
步长 |
输出图像shape |
| 第一层反卷积 |
[batch,2,2,64] |
LeakReLU |
[2,2,1,16] |
[1,2,2,1] |
[batch,4,4,64] |
| 第二层反卷积 |
[batch,4,4,64] |
LeakReLU |
[3,316,32] |
[1,2,2,1] |
[batch,7,7,32] |
| 第三层反卷积 |
[batch,7,7,32] |
LeakReLU |
[3,3,32,64] |
[1,2,2,1] |
[batch,14,14,16] |
| 第四层反卷积 |
[batch,14,14,16] |
LeakReLU |
[3,3,64,64] |
[1,2,2,1] |
[batch,28,28,1] |
3.3 主网络
| 主网络 |
|
|
|
|
| 获取输入 |
占位符 |
类型：tf.float32
|
shape：[batch,28,28,1] |
|
| 前向结构 |
获取卷积输出 |
[batch,2,2,64] |
|
|
|
获取反卷积输出 |
[batch,28,28,1] |
|
|
| 后向结构 |
损失 |
反卷积输出 |
输入数据 |
均值平方差 |
|
优化器 |
tf.train.AdamOptimizer() |
|
|
3.4 训练网络
- MNIST数据集
- 每次从训练集取100张图片
- 因获取图片shape为【100,784】，转换为【100,28,28,1】
- 获取损失
- 每训练20次，输出损失
- 从测试集中取100张图片，输入进网络，获取解码后的图片，并输出
3.5 测试代码
| 测试网络 |
形状：shape |
batch=100 |
| 获取测试集 |
样本：[batch,784] |
|
| 更改样本集的形状 |
样本：[batch,28,28,1] |
|
| 向主网络中传入数据 |
样本：[batch,28,28,1] |
|
| 获取解码器输出数据 |
预测数据：[batch,28,28,1] |
|
| 输出一张真实图片 |
|
|
| 输出一张解码图片 |
|
|
4. 网络结构实现代码：
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import numpy as np
import matplotlib.pyplot as plt
mnist = input_data.read_data_sets(r'E:\PycharmProjects\TensorflowTest\MNIST_data',one_hot=True)
class Convolution:
def __init__(self):
self.filter1 = tf.Variable(tf.truncated_normal([3,3,1,16], stddev=0.1))
self.b1 = tf.Variable(tf.zeros([16]))
self.filter2 = tf.Variable(tf.truncated_normal([3,3,16,32], stddev=0.1))
self.b2 = tf.Variable(tf.zeros([32]))
self.filter3 = tf.Variable(tf.truncated_normal([3,3,32,64], stddev=0.1))
self.b3 = tf.Variable(tf.zeros([64]))
def forward(self,in_x):
conv1 = tf.nn.leaky_relu(tf.add(tf.nn.conv2d(in_x,
self.filter1,
[1,2,2,1],
padding='SAME'), self.b1))
conv2 = tf.nn.leaky_relu(tf.add(tf.nn.conv2d(conv1,
self.filter2,
[1,2,2,1],
padding='SAME'), self.b2))
conv3 = tf.nn.leaky_relu(tf.add(tf.nn.conv2d(conv2,
self.filter3,
[1,2,2,1],
padding='SAME'), self.b3))
return conv3
class Deconvolution:
def __init__(self):
self.filter3 = tf.Variable(tf.truncated_normal([3, 3, 32, 64], stddev=0.1))
self.b3 = tf.Variable(tf.zeros([32]))
self.filter2 = tf.Variable(tf.truncated_normal([3, 3, 16, 32], stddev=0.1))
self.b2 = tf.Variable(tf.zeros([16]))
self.filter1 = tf.Variable(tf.truncated_normal([3, 3, 1, 16], stddev=0.1))
self.b1 = tf.Variable(tf.zeros([1]))
self.strides=[1,2,2,1]
def forward(self,conv_layer):
deconv1 = tf.nn.leaky_relu(tf.nn.conv2d_transpose(conv_layer,
self.filter3,
[100,7,7,32],
strides=self.strides), self.b3)
deconv2 = tf.nn.leaky_relu(tf.nn.conv2d_transpose(deconv1,
self.filter2,
[100,14,14,16],
strides=self.strides), self.b2)
deconv3 = tf.nn.leaky_relu(tf.nn.conv2d_transpose(deconv2,
self.filter1,
[100,28,28,1],
strides=self.strides), self.b1)
return deconv3
class Net:
def __init__(self):
self.conv = Convolution()
self.deconv = Deconvolution()
self.in_x = tf.placeholder(dtype=tf.float32, shape=[None, 28, 28, 1])
self.forward()
self.backward()
def forward(self):
self.conv_layer = self.conv.forward(self.in_x)
self.deconv_layer = self.deconv.forward(self.conv_layer)
def backward(self):
self.loss = tf.reduce_mean((self.deconv_layer-self.in_x)**2)
self.opt = tf.train.AdamOptimizer().minimize(self.loss)
5. 训练网络：获取损失和解码图
if __name__ == '__main__':
net = Net()
with tf.Session() as sess:
init = tf.global_variables_initializer()
sess.run(init)
saver = tf.train.Saver()
plt.ion()
loss_sum = []
for epoch in range(10000):
xs,_ = mnist.train.next_batch(100)
train_xs = np.reshape(xs, [100, 28, 28, 1])
loss, _ = sess.run([net.loss, net.opt],feed_dict={net.in_x: train_xs})
if epoch % 20 ==0:
saver.save(sess,r'E:\PycharmProjects\TensorflowTest\log\CNNAEtrain1.ckpt')
test_xs,_ = mnist.train.next_batch(100)
test_xs = np.reshape(test_xs, [100,28, 28,1])
photo_x = np.reshape(sess.run(net.deconv_layer, feed_dict={net.in_x: test_xs})[1],[28,28])
loss_sum.append(loss)
print('loss: ', loss)
plt.imshow(photo_x)
plt.pause(0.1)
plt.figure("CNN_AE_Loss图")
plt.plot(loss_sum,label='CNN_AE_Loss')
plt.legend()
plt.show()
6. 代码结果输出：
loss: 0.10667889
loss: 0.0795733
loss: 0.051609386
loss: 0.030548936
7. 损失图
8. 效果展示：
原始图片：
生成图片：
9. 测试网络：
测试代码：
f __name__ == '__main__':
net = Net()
with tf.Session() as sess:
init = tf.global_variables_initializer()
sess.run(init)
saver = tf.train.Saver()
saver.restore(sess, r'E:\PycharmProjects\TensorflowTest\log\CNNAEtrain1.ckpt')
test_xs, _ = mnist.train.next_batch(100)
test_xs = np.reshape(test_xs, [100, 28, 28, 1])
photo_x = np.reshape(sess.run(net.deconv_layer, feed_dict={net.in_x: test_xs})[1], [28, 28])
photo_xs = np.reshape(test_xs[1], [28, 28])
plt.figure('原始图片')
plt.imshow(photo_xs)
plt.show()
plt.figure('生成图片')
plt.imshow(photo_x)
plt.show()
原始图片：
生成图片：
