系列文章目录

深度学习GAN（一）之简单介绍

深度学习GAN（二）之基于CIFAR10数据集的例子;

深度学习GAN（三）之基于手写体Mnist数据集的例子;

深度学习GAN（四）之PIX2PIX GAN的例子

---

GAN入门例子

系列文章目录0. 前言1. GAN的简单介绍2. GAN原理3. 如何定义与训练判别器（discriminator model）3.1 定义判别器模型3.2. 数据集归一化已经转化为浮点类型3.3. 随机的取n_sample个真实图片从数据集中3.4. 随机制造n_sample个假的图片3.5.训练判别器4. 如何定义与使用产生器(Generator)4.1 定义产生器4.2 展示假的图片4.3 定义GAN模型4.4. 训练GAN模型5.展示Generator产生的假的图片6 显示一张假图片7. GAN基于CIFAR10数据集的完整代码

0. 前言

CIFAR-10 数据集介绍请看这篇博客

GPU怎么配置请看这篇博客

通过本博客你知道怎么利用GAN产生大量的假的图片、

如下图，左边是真实的图片。右边是利用GAN产生的图片

下图是GAN生成的手写体数字，用了10个epoch

1. GAN的简单介绍

生成式对抗网络（GAN, Generative Adversarial Networks ）是一种深度学习模型，是近年来复杂分布上无监督学习最具前景的方法之一。模型通过框架中（至少）两个模块：生成模型（Generative Model）和判别模型（Discriminative Model）的互相博弈学习产生相当好的输出。原始 GAN 理论中，并不要求 G 和 D 都是神经网络，只需要是能拟合相应生成和判别的函数即可。但实用中一般均使用深度神经网络作为 G 和 D 。一个优秀的GAN应用需要有良好的训练方法，否则可能由于神经网络模型的自由性而导致输出不理想。

要了解GAN的工作原理以及如何在GAN架构中训练深层卷积神经网络模型以生成图像，可能是具有挑战性的。对于初学者来说，一个很好的起点是在计算机视觉领域中使用的标准图像数据集（例如CIFAR小物体照片数据集）上练习开发和使用GAN。使用小型且易于理解的数据集意味着可以快速开发和训练较小的模型，从而可以将重点放在模型体系结构和图像生成过程本身上。

在本博客中，您将发现如何使用具有深度卷积网络的生成对抗网络来生成物体的小照片。

2. GAN原理

在训练过程中，生成器努力地让生成的图像更加真实，而判别器则努力地去识别出图像的真假，这个过程相当与一个二人博弈，随着时间的推移，生成器和判别器在不断地进行对抗。

下图是GAN的整个结构的图片

3. 如何定义与训练判别器（discriminator model）

3.1 定义判别器模型

定义一个判别器模型的函数define_discriminator()，输入参数是图片的大小，默认（32，32，3）

import tensorflow as tfimport tensorflow.keras as kerasimport numpy as npimport matplotlib.pyplot as plt# define the standalone discriminator modeldef define_discriminator(in_shape=(32,32,3)):model = keras.models.Sequential()# normalmodel.add(keras.layers.Conv2D(64, (3,3), padding='same', input_shape=in_shape))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(128, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(128, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(256, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# classifiermodel.add(keras.layers.Flatten())model.add(keras.layers.Dropout(0.4))model.add(keras.layers.Dense(1, activation='sigmoid'))# compile modelopt = keras.optimizers.Adam(lr=0.0002, beta_1=0.5)pile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])return modelif __name__ == '__main__': model = define_discriminator()# summarize the modelmodel.summary()

结果

Model: "sequential"_________________________________________________________________Layer (type) Output Shape Param # =================================================================conv2d (Conv2D) (None, 32, 32, 64) 1792_________________________________________________________________leaky_re_lu (LeakyReLU)(None, 32, 32, 64) 0 _________________________________________________________________conv2d_1 (Conv2D) (None, 16, 16, 128) 73856_________________________________________________________________leaky_re_lu_1 (LeakyReLU) (None, 16, 16, 128) 0 _________________________________________________________________conv2d_2 (Conv2D) (None, 8, 8, 128) 147584 _________________________________________________________________leaky_re_lu_2 (LeakyReLU) (None, 8, 8, 128) 0 _________________________________________________________________conv2d_3 (Conv2D) (None, 4, 4, 256) 295168 _________________________________________________________________leaky_re_lu_3 (LeakyReLU) (None, 4, 4, 256) 0 _________________________________________________________________flatten (Flatten) (None, 4096) 0 _________________________________________________________________dropout (Dropout) (None, 4096) 0 _________________________________________________________________dense (Dense)(None, 1) 4097=================================================================Total params: 522,497Trainable params: 522,497Non-trainable params: 0

3.2. 数据集归一化已经转化为浮点类型

把图片的像素点从[0,255]转成[-1, 1]

# load and prepare cifar10 training imagesdef load_real_samples():# load cifar10 dataset(trainX, _), (_, _) = tf.keras.datasets.cifar10.load_data()# convert from unsigned ints to floatsX = trainX.astype('float32')# scale from [0,255] to [-1,1]X = (X - 127.5) / 127.5return X

3.3. 随机的取n_sample个真实图片从数据集中

Y 都是1

# select real samplesdef generate_real_samples(dataset, n_samples):# choose random instancesix = np.random.randint(0, dataset.shape[0], n_samples)# retrieve selected imagesX = dataset[ix]# generate 'real' class labels (1)y = np.ones((n_samples, 1))return X, y

3.4. 随机制造n_sample个假的图片

Y都是0

def generate_fake_samples1(n_samples):# generate uniform random numbers in [0,1]X = np.random.rand(32 * 32 * 3 * n_samples)# update to have the range [-1, 1]X = -1 + X * 2# reshape into a batch of color imagesX = X.reshape((n_samples, 32, 32, 3))# generate 'fake' class labels (0)y = np.zeros((n_samples, 1))return X, y

3.5.训练判别器

真实图片与假的图片一起训练。

# train the discriminator modeldef train_discriminator(model, dataset, n_iter=20, n_batch=128):half_batch = int(n_batch / 2)# manually enumerate epochsfor i in range(n_iter):# get randomly selected 'real' samplesX_real, y_real = generate_real_samples(dataset, half_batch)# update discriminator on real samples_, real_acc = model.train_on_batch(X_real, y_real)# generate 'fake' examplesX_fake, y_fake = generate_fake_samples1(half_batch)# update discriminator on fake samples_, fake_acc = model.train_on_batch(X_fake, y_fake)# summarize performanceprint('>%d real=%.0f%% fake=%.0f%%' % (i+1, real_acc*100, fake_acc*100))def test_train_discriminator():# define the discriminator modelmodel = define_discriminator()# load image datadataset = load_real_samples()# fit the modeltrain_discriminator(model, dataset)if __name__ == '__main__':test_train_discriminator()

结果如下，可以看出训练的时候真实的图片的准确率与假的图片的准确率

>1 real=83% fake=0%>2 real=100% fake=0%>3 real=100% fake=8%>4 real=94% fake=31%>5 real=97% fake=62%>6 real=86% fake=92%>7 real=98% fake=98%>8 real=92% fake=100%>9 real=88% fake=100%>10 real=94% fake=100%>11 real=94% fake=100%>12 real=100% fake=100%>13 real=91% fake=100%>14 real=97% fake=100%>15 real=98% fake=100%>16 real=100% fake=100%>17 real=97% fake=100%>18 real=100% fake=100%>19 real=100% fake=100%>20 real=100% fake=100%

4. 如何定义与使用产生器(Generator)

4.1 定义产生器

产生器的生成正好和判别器是相反的。

是由100维的向量反向生成（32，32，3）的图片

此处的Generator不用指定loss函数与优化器，因为它不能直接训练，

下面还有其它的。

# define the standalone generator modeldef define_generator(latent_dim):model = keras.models.Sequential()# foundation for 4x4 imagen_nodes = 256 * 4 * 4model.add(keras.layers.Dense(n_nodes, input_dim=latent_dim))model.add(keras.layers.LeakyReLU(alpha=0.2))model.add(keras.layers.Reshape((4, 4, 256)))# upsample to 8x8model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# upsample to 16x16model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# upsample to 32x32model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# output layermodel.add(keras.layers.Conv2D(3, (3,3), activation='tanh', padding='same'))return model# define the size of the latent spacelatent_dim = 100# define the generator modelmodel = define_generator(latent_dim)# summarize the modelmodel.summary()

执行结果如下，

Model: "sequential"_________________________________________________________________Layer (type) Output Shape Param # =================================================================dense (Dense)(None, 4096) 413696 _________________________________________________________________leaky_re_lu (LeakyReLU)(None, 4096) 0 _________________________________________________________________reshape (Reshape) (None, 4, 4, 256) 0 _________________________________________________________________conv2d_transpose (Conv2DTran (None, 8, 8, 128) 524416 _________________________________________________________________leaky_re_lu_1 (LeakyReLU) (None, 8, 8, 128) 0 _________________________________________________________________conv2d_transpose_1 (Conv2DTr (None, 16, 16, 128) 262272 _________________________________________________________________leaky_re_lu_2 (LeakyReLU) (None, 16, 16, 128) 0 _________________________________________________________________conv2d_transpose_2 (Conv2DTr (None, 32, 32, 128) 262272 _________________________________________________________________leaky_re_lu_3 (LeakyReLU) (None, 32, 32, 128) 0 _________________________________________________________________conv2d (Conv2D) (None, 32, 32, 3) 3459=================================================================Total params: 1,466,115Trainable params: 1,466,115Non-trainable params: 0

4.2 展示假的图片

generate_latent_points（）产生n_samples个随机向量，每个向量默认为100维

generate_fake_samples（）根据产生器生成假的图片

show_fake_sample（）展示假的图片

# generate points in latent space as input for the generatordef generate_latent_points(latent_dim, n_samples):# generate points in the latent spacex_input = np.random.randn(latent_dim * n_samples)# reshape into a batch of inputs for the networkx_input = x_input.reshape(n_samples, latent_dim)return x_input# use the generator to generate n fake examples, with class labelsdef generate_fake_samples(g_model, latent_dim, n_samples):# generate points in latent spacex_input = generate_latent_points(latent_dim, n_samples)# predict outputsX = g_model.predict(x_input)# create 'fake' class labels (0)y = np.zeros((n_samples, 1))return X, ydef show_fake_sample():# size of the latent spacelatent_dim = 100# define the discriminator modelmodel = define_generator(latent_dim)# generate samplesn_samples = 49X, _ = generate_fake_samples(model, latent_dim, n_samples)# scale pixel values from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the generated samplesfor i in range(n_samples):# define subplotplt.subplot(7, 7, 1 + i)# turn off axis labelsplt.axis('off')# plot single imageplt.imshow(X[i])# show the figureplt.show()if __name__ == '__main__':show_fake_sample()

执行结果如下，因为还没有训练，所以图片都是灰色的。

4.3 定义GAN模型

Gan的模型其实是由Generator与Desciminator组成的

# define the combined generator and discriminator model, for updating the generatordef define_gan(g_model, d_model):# make weights in the discriminator not trainabled_model.trainable = False# connect themmodel = tf.keras.models.Sequential()# add generatormodel.add(g_model)# add the discriminatormodel.add(d_model)# compile modelopt = tf.keras.optimizers.Adam(lr=0.0002, beta_1=0.5)pile(loss='binary_crossentropy', optimizer=opt)return modeldef show_gan_module():# size of the latent spacelatent_dim = 100# create the discriminatord_model = define_discriminator()# create the generatorg_model = define_generator(latent_dim)# create the gangan_model = define_gan(g_model, d_model)# summarize gan modelgan_model.summary()if __name__ == '__main__':show_gan_module()

执行结果如下：

Model: "sequential_2"_________________________________________________________________Layer (type) Output Shape Param # =================================================================sequential_1 (Sequential) (None, 32, 32, 3) 1466115 _________________________________________________________________sequential (Sequential)(None, 1) 522497 =================================================================Total params: 1,988,612Trainable params: 1,466,115Non-trainable params: 522,497

4.4. 训练GAN模型

# evaluate the discriminator, plot generated images, save generator modeldef summarize_performance(epoch, g_model, d_model, dataset, latent_dim, n_samples=150):# prepare real samplesX_real, y_real = generate_real_samples(dataset, n_samples)# evaluate discriminator on real examples_, acc_real = d_model.evaluate(X_real, y_real, verbose=0)# prepare fake examplesx_fake, y_fake = generate_fake_samples(g_model, latent_dim, n_samples)# evaluate discriminator on fake examples_, acc_fake = d_model.evaluate(x_fake, y_fake, verbose=0)# summarize discriminator performanceprint('>Accuracy real: %.0f%%, fake: %.0f%%' % (acc_real * 100, acc_fake * 100))# save plot#save_plot(x_fake, epoch)# save the generator model tile filefilename = 'generator_model_%03d.h5' % (epoch + 1)g_model.save(filename)# train the generator and discriminatordef train(g_model, d_model, gan_model, dataset, latent_dim, n_epochs=200, n_batch=128):bat_per_epo = int(dataset.shape[0] / n_batch)half_batch = int(n_batch / 2)# manually enumerate epochsfor i in range(n_epochs):# enumerate batches over the training setfor j in range(bat_per_epo):# get randomly selected 'real' samplesX_real, y_real = generate_real_samples(dataset, half_batch)# update discriminator model weightsd_loss1, _ = d_model.train_on_batch(X_real, y_real)# generate 'fake' examplesX_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch)# update discriminator model weightsd_loss2, _ = d_model.train_on_batch(X_fake, y_fake)# prepare points in latent space as input for the generatorX_gan = generate_latent_points(latent_dim, n_batch)# create inverted labels for the fake samplesy_gan = np.ones((n_batch, 1))# update the generator via the discriminator's errorg_loss = gan_model.train_on_batch(X_gan, y_gan)# summarize loss on this batchprint('>%d, %d/%d, d1=%.3f, d2=%.3f g=%.3f' %(i + 1, j + 1, bat_per_epo, d_loss1, d_loss2, g_loss))# evaluate the model performance, sometimesif (i + 1) % 10 == 0:summarize_performance(i, g_model, d_model, dataset, latent_dim)def train_gan():# size of the latent spacelatent_dim = 100# create the discriminatord_model = define_discriminator()# create the generatorg_model = define_generator(latent_dim)# create the gangan_model = define_gan(g_model, d_model)# load image datadataset = load_real_samples()# train modeltrain(g_model, d_model, gan_model, dataset, latent_dim)if __name__ == '__main__':train_gan()

5.展示Generator产生的假的图片

加载训练好的权重参数，随机产生100个100维的点。利用Generator产生假图片并展示出来

注意：模型的权重参数文件名字有可能要改变一下。

def show_imgs_for_final_generator_model():# load modelmodel = tf.keras.models.load_model('generator_model_010.h5')# generate imageslatent_points = generate_latent_points(100, 100)# generate imagesX = model.predict(latent_points)# scale from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the resultcreate_plot(X, 10)

10个epoch之后，结果如下：感觉虽然每个图片有都有不同。但是还是很难分别它是什么。

90，100个epoch之后。我们就可以识别出具体的图片是什么了。

100 epoch之后， 就稳定了

6 显示一张假图片

利用一个所有值都是0.75的向量产生一个图片

def show_single_imgs():model = tf.keras.models.load_model('generator_model_200.h5')# all 0svector = np.asarray([[0.75 for _ in range(100)]])# generate imageX = model.predict(vector)# scale from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the resultplt.imshow(X[0, :, :])plt.show()

结果如下，看起来像鹿，又像是鹿与马组成的动物。

7. GAN基于CIFAR10数据集的完整代码

import tensorflow as tfimport tensorflow.keras as kerasimport numpy as npimport matplotlib.pyplot as plt# define the standalone discriminator modeldef define_discriminator(in_shape=(32,32,3)):model = keras.models.Sequential()# normalmodel.add(keras.layers.Conv2D(64, (3,3), padding='same', input_shape=in_shape))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(128, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(128, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# downsamplemodel.add(keras.layers.Conv2D(256, (3,3), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# classifiermodel.add(keras.layers.Flatten())model.add(keras.layers.Dropout(0.4))model.add(keras.layers.Dense(1, activation='sigmoid'))# compile modelopt = keras.optimizers.Adam(lr=0.0002, beta_1=0.5)pile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])model.summary()return model# load and prepare cifar10 training imagesdef load_real_samples():# load cifar10 dataset(trainX, _), (_, _) = tf.keras.datasets.cifar10.load_data()# convert from unsigned ints to floatsX = trainX.astype('float32')# scale from [0,255] to [-1,1]X = (X - 127.5) / 127.5return X# select real samplesdef generate_real_samples(dataset, n_samples):# choose random instancesix = np.random.randint(0, dataset.shape[0], n_samples)# retrieve selected imagesX = dataset[ix]# generate 'real' class labels (1)y = np.ones((n_samples, 1))return X, ydef generate_fake_samples1(n_samples):# generate uniform random numbers in [0,1]X = np.random.rand(32 * 32 * 3 * n_samples)# update to have the range [-1, 1]X = -1 + X * 2# reshape into a batch of color imagesX = X.reshape((n_samples, 32, 32, 3))# generate 'fake' class labels (0)y = np.zeros((n_samples, 1))return X, y# train the discriminator modeldef train_discriminator(model, dataset, n_iter=20, n_batch=128):half_batch = int(n_batch / 2)# manually enumerate epochsfor i in range(n_iter):# get randomly selected 'real' samplesX_real, y_real = generate_real_samples(dataset, half_batch)# update discriminator on real samples_, real_acc = model.train_on_batch(X_real, y_real)# generate 'fake' examplesX_fake, y_fake = generate_fake_samples1(half_batch)# update discriminator on fake samples_, fake_acc = model.train_on_batch(X_fake, y_fake)# summarize performanceprint('>%d real=%.0f%% fake=%.0f%%' % (i+1, real_acc*100, fake_acc*100))def test_train_discriminator():# define the discriminator modelmodel = define_discriminator()# load image datadataset = load_real_samples()# fit the modeltrain_discriminator(model, dataset)# define the standalone generator modeldef define_generator(latent_dim):model = keras.models.Sequential()# foundation for 4x4 imagen_nodes = 256 * 4 * 4model.add(keras.layers.Dense(n_nodes, input_dim=latent_dim))model.add(keras.layers.LeakyReLU(alpha=0.2))model.add(keras.layers.Reshape((4, 4, 256)))# upsample to 8x8model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# upsample to 16x16model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# upsample to 32x32model.add(keras.layers.Conv2DTranspose(128, (4,4), strides=(2,2), padding='same'))model.add(keras.layers.LeakyReLU(alpha=0.2))# output layermodel.add(keras.layers.Conv2D(3, (3,3), activation='tanh', padding='same'))return model# generate points in latent space as input for the generatordef generate_latent_points(latent_dim, n_samples):# generate points in the latent spacex_input = np.random.randn(latent_dim * n_samples)# reshape into a batch of inputs for the networkx_input = x_input.reshape(n_samples, latent_dim)return x_input# use the generator to generate n fake examples, with class labelsdef generate_fake_samples(g_model, latent_dim, n_samples):# generate points in latent spacex_input = generate_latent_points(latent_dim, n_samples)# predict outputsX = g_model.predict(x_input)# create 'fake' class labels (0)y = np.zeros((n_samples, 1))return X, ydef show_fake_sample():# size of the latent spacelatent_dim = 100# define the discriminator modelmodel = define_generator(latent_dim)# generate samplesn_samples = 49X, _ = generate_fake_samples(model, latent_dim, n_samples)# scale pixel values from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the generated samplesfor i in range(n_samples):# define subplotplt.subplot(7, 7, 1 + i)# turn off axis labelsplt.axis('off')# plot single imageplt.imshow(X[i])# show the figureplt.show()# define the combined generator and discriminator model, for updating the generatordef define_gan(g_model, d_model):# make weights in the discriminator not trainabled_model.trainable = False# connect themmodel = tf.keras.models.Sequential()# add generatormodel.add(g_model)# add the discriminatormodel.add(d_model)# compile modelopt = tf.keras.optimizers.Adam(lr=0.0002, beta_1=0.5)pile(loss='binary_crossentropy', optimizer=opt)return modeldef show_gan_module():# size of the latent spacelatent_dim = 100# create the discriminatord_model = define_discriminator()# create the generatorg_model = define_generator(latent_dim)# create the gangan_model = define_gan(g_model, d_model)# summarize gan modelgan_model.summary()# evaluate the discriminator, plot generated images, save generator modeldef summarize_performance(epoch, g_model, d_model, dataset, latent_dim, n_samples=150):# prepare real samplesX_real, y_real = generate_real_samples(dataset, n_samples)# evaluate discriminator on real examples_, acc_real = d_model.evaluate(X_real, y_real, verbose=0)# prepare fake examplesx_fake, y_fake = generate_fake_samples(g_model, latent_dim, n_samples)# evaluate discriminator on fake examples_, acc_fake = d_model.evaluate(x_fake, y_fake, verbose=0)# summarize discriminator performanceprint('>Accuracy real: %.0f%%, fake: %.0f%%' % (acc_real * 100, acc_fake * 100))# save plot#save_plot(x_fake, epoch)# save the generator model tile filefilename = 'generator_model_%03d.h5' % (epoch + 1)g_model.save(filename)# train the generator and discriminatordef train(g_model, d_model, gan_model, dataset, latent_dim, n_epochs=200, n_batch=128):bat_per_epo = int(dataset.shape[0] / n_batch)half_batch = int(n_batch / 2)# manually enumerate epochsfor i in range(n_epochs):# enumerate batches over the training setfor j in range(bat_per_epo):# get randomly selected 'real' samplesX_real, y_real = generate_real_samples(dataset, half_batch)# update discriminator model weightsd_loss1, _ = d_model.train_on_batch(X_real, y_real)# generate 'fake' examplesX_fake, y_fake = generate_fake_samples(g_model, latent_dim, half_batch)# update discriminator model weightsd_loss2, _ = d_model.train_on_batch(X_fake, y_fake)# prepare points in latent space as input for the generatorX_gan = generate_latent_points(latent_dim, n_batch)# create inverted labels for the fake samplesy_gan = np.ones((n_batch, 1))# update the generator via the discriminator's errorg_loss = gan_model.train_on_batch(X_gan, y_gan)# summarize loss on this batchprint('>%d, %d/%d, d1=%.3f, d2=%.3f g=%.3f' %(i + 1, j + 1, bat_per_epo, d_loss1, d_loss2, g_loss))# evaluate the model performance, sometimesif (i + 1) % 10 == 0:summarize_performance(i, g_model, d_model, dataset, latent_dim)def train_gan():# size of the latent spacelatent_dim = 100# create the discriminatord_model = define_discriminator()# create the generatorg_model = define_generator(latent_dim)# create the gangan_model = define_gan(g_model, d_model)# load image datadataset = load_real_samples()# train modeltrain(g_model, d_model, gan_model, dataset, latent_dim)# generate points in latent space as input for the generatordef generate_latent_points(latent_dim, n_samples):# generate points in the latent spacex_input = np.random.randn(latent_dim * n_samples)# reshape into a batch of inputs for the networkx_input = x_input.reshape(n_samples, latent_dim)return x_input# plot the generated imagesdef create_plot(examples, n):# plot imagesfor i in range(n * n):# define subplotplt.subplot(n, n, 1 + i)# turn off axisplt.axis('off')# plot raw pixel dataplt.imshow(examples[i, :, :])plt.show()def show_imgs_for_final_generator_model():# load modelmodel = tf.keras.models.load_model('generator_model_010.h5')# generate imageslatent_points = generate_latent_points(100, 100)# generate imagesX = model.predict(latent_points)# scale from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the resultcreate_plot(X, 10)def show_single_imgs():model = tf.keras.models.load_model('generator_model_200.h5')# all 0svector = np.asarray([[0.75 for _ in range(100)]])# generate imageX = model.predict(vector)# scale from [-1,1] to [0,1]X = (X + 1) / 2.0# plot the resultplt.imshow(X[0, :, :])plt.show()if __name__ == '__main__':#define_discriminator()#test_train_discriminator()# show_fake_sample()#show_gan_module()train_gan()show_imgs_for_final_generator_model()

如果觉得《[深度学习-实践]GAN入门例子-利用Tensorflow Keras与数据集CIFAR10生成新图片》对你有帮助，请点赞、收藏，并留下你的观点哦！