該篇文章與上篇文章內(nèi)容相差不多，但是主要的網(wǎng)絡(luò)結(jié)構(gòu)不同，上篇文章采用的是GAN網(wǎng)絡(luò)結(jié)構(gòu)，而這篇文章采用的是DCGAN網(wǎng)絡(luò)結(jié)構(gòu)。兩者的差異在于以下幾點(diǎn)：

（1）使用卷積和去卷積代替池化層。
（2）在生成器和判別器中都添加了批量歸一化操作。
（3）去掉了全連接層，使用全局池化層替代。
（4）生成器的輸出層使用Tanh 激活函數(shù)，其他層使用RELU。
（5）判別器的所有層都是用LeakyReLU 激活函數(shù)。
其中最本質(zhì)的一點(diǎn)就是使用了卷積和去卷積。

1.導(dǎo)入庫(kù)

import numpy as np import glob,imageio,os,PIL import matplotlib.pyplot as plt import tensorflow as tf from tensorflow.keras import layers

2.數(shù)據(jù)準(zhǔn)備

歸一化-打亂-batch

(train_images,train_labels),(_,_) = tf.keras.datasets.mnist.load_data() train_images = train_images.reshape(train_images.shape[0],28,28,1).astype('float32') train_images = (train_images - 127.5) / 127.5#歸一化到[-1,1]之間 batch_size = 256 buffer_size = 60000 datasets = tf.data.Dataset.from_tensor_slices(train_images).shuffle(buffer_size).batch(batch_size)

3.生成器與判別器的構(gòu)建

生成器的構(gòu)建中，采用了tf.keras.layers.Conv2DTranspose，也就是去卷積，它的目的是將經(jīng)過(guò)池化層以后縮小的矩陣擴(kuò)大到一定的大小，比如說(shuō)從3 * 3 擴(kuò)大到 5 * 5，如下圖所示：

def Generator_model():model = tf.keras.Sequential([tf.keras.layers.Dense(7*7*256,use_bias=False,input_shape=(100,)),#輸入一個(gè)長(zhǎng)度為100的隨機(jī)向量，tf.keras.layers.BatchNormalization(),tf.keras.layers.LeakyReLU(),tf.keras.layers.Reshape((7,7,256)),#調(diào)整為形狀為（7,7,256)。tf.keras.layers.Conv2DTranspose(128,(5,5),strides=(1,1),padding="same",use_bias=False),tf.keras.layers.BatchNormalization(),tf.keras.layers.LeakyReLU(),tf.keras.layers.Conv2DTranspose(64,(5,5),strides=(2,2),padding="same",use_bias=False),tf.keras.layers.BatchNormalization(),tf.keras.layers.LeakyReLU(),tf.keras.layers.Conv2DTranspose(1,(5,5),strides=(2,2),padding="same",use_bias=False,activation='tanh')#激活函數(shù)采用tanh函數(shù)])return model generator = Generator_model() generator.summary()

網(wǎng)絡(luò)結(jié)構(gòu)如下所示：

Model: "sequential" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= dense (Dense) (None, 12544) 1254400 _________________________________________________________________ batch_normalization (BatchNo (None, 12544) 50176 _________________________________________________________________ leaky_re_lu (LeakyReLU) (None, 12544) 0 _________________________________________________________________ reshape (Reshape) (None, 7, 7, 256) 0 _________________________________________________________________ conv2d_transpose (Conv2DTran (None, 7, 7, 128) 819200 _________________________________________________________________ batch_normalization_1 (Batch (None, 7, 7, 128) 512 _________________________________________________________________ leaky_re_lu_1 (LeakyReLU) (None, 7, 7, 128) 0 _________________________________________________________________ conv2d_transpose_1 (Conv2DTr (None, 14, 14, 64) 204800 _________________________________________________________________ batch_normalization_2 (Batch (None, 14, 14, 64) 256 _________________________________________________________________ leaky_re_lu_2 (LeakyReLU) (None, 14, 14, 64) 0 _________________________________________________________________ conv2d_transpose_2 (Conv2DTr (None, 28, 28, 1) 1600 ================================================================= Total params: 2,330,944 Trainable params: 2,305,472 Non-trainable params: 25,472 _________________________________________________________________

判別器的構(gòu)建：

def Disciminator_model():model = tf.keras.Sequential([tf.keras.layers.Conv2D(64,(5,5),strides=(2,2),padding="same",input_shape=(28,28,1)),tf.keras.layers.LeakyReLU(),tf.keras.layers.Dropout(0.3),tf.keras.layers.Conv2D(128,(5,5),strides=(2,2),padding="same"),tf.keras.layers.LeakyReLU(),tf.keras.layers.Dropout(0.3),tf.keras.layers.Flatten(),tf.keras.layers.Dense(1,activation='sigmoid')])return model discriminator = Disciminator_model()

網(wǎng)絡(luò)結(jié)構(gòu)如下所示：

Model: "sequential_1" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= conv2d (Conv2D) (None, 14, 14, 64) 1664 _________________________________________________________________ leaky_re_lu_3 (LeakyReLU) (None, 14, 14, 64) 0 _________________________________________________________________ dropout (Dropout) (None, 14, 14, 64) 0 _________________________________________________________________ conv2d_1 (Conv2D) (None, 7, 7, 128) 204928 _________________________________________________________________ leaky_re_lu_4 (LeakyReLU) (None, 7, 7, 128) 0 _________________________________________________________________ dropout_1 (Dropout) (None, 7, 7, 128) 0 _________________________________________________________________ flatten (Flatten) (None, 6272) 0 _________________________________________________________________ dense_1 (Dense) (None, 1) 6273 ================================================================= Total params: 212,865 Trainable params: 212,865 Non-trainable params: 0 _________________________________________________________________

這一部分是與上篇文章代碼的本質(zhì)差異。

4.其余操作

由于硬件原因，epochs設(shè)置的是60。如果有不明白的地方，請(qǐng)參考上篇文章

#loss值 cross_entropy = tf.keras.losses.BinaryCrossentropy(from_logits=True) #Dis_loss def Discriminator_loss(real_out,fake_out):real_loss = cross_entropy(tf.ones_like(real_out),real_out)fake_loss = cross_entropy(tf.zeros_like(fake_out),fake_out)return real_loss+fake_loss #Gen_loss def Generator_loss(fake_out):return cross_entropy(tf.ones_like(fake_out),fake_out) #Dis_opt generator_optimizer = tf.keras.optimizers.Adam(1e-4) discriminator_optimizer = tf.keras.optimizers.Adam(1e-4)epochs = 60 noise_dim = 100 num_exp_to_generate = 16seed = tf.random.normal([num_exp_to_generate,noise_dim])def train_step(images):noise = tf.random.normal([batch_size,noise_dim])with tf.GradientTape() as gen_tape,tf.GradientTape() as dis_tape:gen_images = generator(noise,training = True)real_out = discriminator(images,training = True)fake_out = discriminator(gen_images,training = True)gen_loss = Generator_loss(fake_out)dis_loss = Discriminator_loss(real_out,fake_out)gen_gradient = gen_tape.gradient(gen_loss,generator.trainable_variables)dis_gradient = dis_tape.gradient(dis_loss,discriminator.trainable_variables)generator_optimizer.apply_gradients(zip(gen_gradient,generator.trainable_variables))discriminator_optimizer.apply_gradients(zip(dis_gradient,discriminator.trainable_variables))def Generator_plot_image(gen_model,test_noise,epoch):pre_images = gen_model(test_noise,training = False)#根據(jù)test_noise生成圖片，生成器設(shè)置為不可訓(xùn)練fig = plt.figure(figsize=(4,4))for i in range(pre_images.shape[0]):plt.subplot(4,4,i+1)plt.imshow((pre_images[i,:,:,0]+1)/2,cmap='gray')#之前歸一化為[-1,1]之間，現(xiàn)在+1然后除以2，使之在[0,1]之間plt.axis('off')fig.savefig("E:/tmp/.keras/datasets/num_gen_DCGAN/%05d.png" % epoch)plt.close()def train(dataset,epochs):for epoch in range(epochs):for img_batch in dataset:train_step(img_batch)print('.',end = '')print()Generator_plot_image(generator, seed, epoch) train(datasets,epochs)

產(chǎn)生的結(jié)果如下所示（服務(wù)器到期了，然后沒(méi)有運(yùn)行完，這是效果比較好的一個(gè)圖片）：

努力加油a啊

創(chuàng)作挑戰(zhàn)賽新人創(chuàng)作獎(jiǎng)勵(lì)來(lái)咯，堅(jiān)持創(chuàng)作打卡瓜分現(xiàn)金大獎(jiǎng)

總結(jié)

以上是生活随笔為你收集整理的深度学习之基于DCGAN实现手写数字生成的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。