吴裕雄--天生自然神经网络与深度学习实战Python+Keras+TensorFlow：自动编解码器网络的原理与实现
from keras.layers import Dense, Input from keras.layers import Conv2D, Flatten from keras.layers import Reshape, Conv2DTranspose from keras.models import Model from keras.datasets import mnist from keras import backend as K import numpy as np import matplotlib.pyplot as plt #加载手写数字图片数据 (x_train, _), (x_test, _) = mnist.load_data() image_size = x_train.shape[1] #把图片大小统一转换成28*28,并把像素点值都转换为[0,1]之间 x_train = np.reshape(x_train, [-1, image_size, image_size, 1]) x_test = np.reshape(x_test, [-1, image_size, image_size, 1]) x_train = x_train.astype(\'float32\') / 255 x_test = x_test.astype(\'float32\') / 255 input_shape = (image_size, image_size, 1) batch_size = 32 #对图片做3*3分割 kernel_size = 3 #让编码器将输入图片编码成含有16个元素的向量 latent_dim = 16 inputs = Input(shape=input_shape, name=\'encoder_input\') x = inputs \'\'\' 编码器含有两个卷积层，第一个卷积层将每个3*3切片计算成含有32个元素的向量，第二个卷积层将3*3切片 计算成含有64个元素的向量 \'\'\' layer_filters = [32, 64] for filters in layer_filters: #stride=2表明每次挪到2个像素，如此一来做一次卷积运算后输出大小会减半 x = Conv2D(filters = filters, kernel_size = kernel_size, activation=\'relu\', strides = 2, padding = \'same\')(x) shape = K.int_shape(x) print(\'shape: \', shape) print(shape[1]) x = Flatten()(x) #最后一层全连接网络输出含有16个元素的向量 latent = Dense(latent_dim, name = \'latent_vector\')(x) encoder = Model(inputs, latent, name=\'encoder\') encoder.summary()
#构造解码器，解码器的输入正好是编码器的输出结果 latent_inputs = Input(shape = (latent_dim, ), name = \'decoder_input\') \'\'\' 它的结构正好和编码器相反，它先是一个全连接层，然后是两层反卷积网络 \'\'\' x = Dense(shape[1] * shape[2] * shape[3])(latent_inputs) x = Reshape((shape[1], shape[2], shape[3]))(x) #两层与编码器对应的反卷积网络 for filters in layer_filters[::-1]: x = Conv2DTranspose(filters = filters, kernel_size = kernel_size, activation=\'relu\', strides = 2, padding = \'same\')(x) outputs = Conv2DTranspose(filters = 1, kernel_size = kernel_size, activation = \'sigmoid\', padding = \'same\', name = \'decoder_output\')(x) decoder = Model(latent_inputs, outputs, name = \'decoder\') decoder.summary()
autoencoder = Model(inputs, decoder(encoder(inputs)), name = \'autoencoder\') autoencoder.summary() \'\'\' 网络训练时，我们采用最小和方差,也就是我们希望解码器输出的图片与输入编码器的图片，在像素上的差异 尽可能的小 \'\'\' autoencoder.compile(loss=\'mse\', optimizer=\'adam\') autoencoder.fit(x_train, x_train, validation_data=(x_test, x_test), epochs = 1, batch_size = batch_size) \'\'\' x_test是输入编码器的测试图片,我们看看解码器输出的图片与输入时是否差别不大 \'\'\' x_decoded = autoencoder.predict(x_test) #把测试图片集中的前8张显示出来，看看解码器生成的图片是否与原图片足够相似 imgs = np.concatenate([x_test[:8], x_decoded[: 8]]) imgs = imgs.reshape((4, 4, image_size, image_size)) imgs = np.vstack([np.hstack(i) for i in imgs]) plt.figure() plt.axis(\'off\') plt.title(\'Input: 1st 2 rows, Decoded: last 2 rows\') plt.imshow(imgs, interpolation=\'none\', cmap=\'gray\') plt.show()