Ian J.Goodfellow、JeanIan J.Goodfellow、Jean

生成式对抗网络(gennerative adversarial
network,GAN),谷歌2014年提出网络模型。灵感自二人口博弈的零和博弈,目前极度恼火的非监督深度上。GAN之父,Ian
J.Goodfellow,公认人工智能顶级专家。

生成式对抗网络(gennerative adversarial
network,GAN),谷歌2014年提出网络模型。灵感自二口博弈的零和博弈,目前极端恼火的非监督深度上。GAN之父,Ian
J.Goodfellow,公认人工智能顶级专家。

原理。
生成式对为网络包含一个别模型(generative
model,G)和一个分辨模型(discriminative model,D)。Ian J.Goodfellow、Jean
Pouget-Abadie、Mehdi Mirza、Bing Xu、David Warde-Farley、Sherjil
Ozair、Aaron Courville、Yoshua Bengio论文,《Generative Adversarial
Network》,https://arxiv.org/abs/1406.2661 。
生成式对抗网络布局:
噪音数据->生成模型->假图片—|
|->判别模型->真/假
打乱训练多少->训练集->真图片-|
生成式对抗网络重大解决什么由训练样本中学习出新样本。生成模型负责训练有样本的布,如果训练样本是图表就转变相似的图纸,如果训练样本是文章名子就变更相似的篇章名子。判别模型是一个二分类器,用来判定输入样本是真实数据或者训练转变的范本。
生成式对抗网络优化,是一个次初次极小巨博弈(minimax two-player
game)问题。使生成模型输出在输入被判别模型时,判断模型秀难判断是实事求是数据还是虚似数据。训练好之转模型,能拿一个噪音向量转化成为和教练集类似之范本。Argustus
Odena、Christopher Olah、Jonathon Shlens论文《Coditional Image Synthesis
with Auxiliary Classifier GANs》。
扶持分类器生成式对抗网络(auxiliary classifier GAN,AC-GAN)实现。

原理。
生成式对施网络包含一个变迁模型(generative
model,G)和一个分辨模型(discriminative model,D)。Ian J.Goodfellow、Jean
Pouget-Abadie、Mehdi Mirza、Bing Xu、David Warde-Farley、Sherjil
Ozair、Aaron Courville、Yoshua Bengio论文,《Generative Adversarial
Network》,https://arxiv.org/abs/1406.2661 。
生成式对抗网络布局:
噪音数据->生成模型->假图片—|
|->判别模型->真/假
打乱训练多少->训练集->真图片-|
生成式对抗网络重大解决什么从训练样本中学习出新样本。生成模型负责训练有样本的遍布,如果训练样本是图表就变化相似的图纸,如果训练样本是文章名子就变相似的篇章名子。判别模型是一个二分类器,用来判定输入样本是实际数据还是教练转变的样书。
生成式对抗网络优化,是一个亚首批极小巨博弈(minimax two-player
game)问题。使生成模型输出在输入被判别模型时,判断模型秀难判断是真数据要虚似数据。训练好之变动模型,能拿一个噪音向量转化成为与教练集类似之样书。Argustus
Odena、Christopher Olah、Jonathon Shlens论文《Coditional Image Synthesis
with Auxiliary Classifier GANs》。
帮分类器生成式对抗网络(auxiliary classifier GAN,AC-GAN)实现。

生成式对抗网络使用。生成数字,生成人脸图像。

生成式对抗网络采用。生成数字,生成人脸图像。

生成式对抗网络实现。https://github.com/fchollet/keras/blob/master/examples/mnist\_acgan.py

Augustus Odena、Chistopher Olah和Jonathon Shlens 论文《Conditional Image
Synthesis With Auxiliary Classifier GANs》。
通过噪声,让别模型G生成虚假数据,和诚数据并送至判别模型D,判别模型一方面输出数据真/假,一方面输出图片分类。
首先定义生成模型,目的是充分成一对(z,L)数据,z是噪声向量,L是(1,28,28)的图像空间。

生成式对抗网络实现。https://github.com/fchollet/keras/blob/master/examples/mnist\_acgan.py

Augustus Odena、Chistopher Olah和Jonathon Shlens 论文《Conditional Image
Synthesis With Auxiliary Classifier GANs》。
透过噪声,让别模型G生成虚假数据,和忠实数据并送及判别模型D,判别模型一方面输出数据真/假,一方面输出图片分类。
首先定义生成模型,目的是殊成一对(z,L)数据,z是噪声向量,L是(1,28,28)的图像空间。

def build_generator(latent_size):
cnn = Sequential()
cnn.add(Dense(1024, input_dim=latent_size, activation=’relu’))
cnn.add(Dense(128 * 7 * 7, activation=’relu’))
cnn.add(Reshape((128, 7, 7)))
#高达采样,图你尺寸变为 14X14
cnn.add(UpSampling2D(size=(2,2)))
cnn.add(Convolution2D(256, 5, 5, border_mode=’same’, activation=’relu’,
init=’glorot_normal’))
#达到采样,图像尺寸变为28X28
cnn.add(UpSampling2D(size=(2,2)))
cnn.add(Convolution2D(128, 5, 5, border_mode=’same’, activation=’relu’,
init=’glorot_normal’))
#规约到1个通道
cnn.add(Convolution2D(1, 2, 2, border_mode=’same’, activation=’tanh’,
init=’glorot_normal’))
#变模型输入层,特征向量
latent = Input(shape=(latent_size, ))
#转变模型输入层,标记
image_class = Input(shape=(1,), dtype=’int32′)
cls = Flatten()(Embedding(10, latent_size,
init=’glorot_normal’)(image_class))
h = merge([latent, cls], mode=’mul’)
fake_image = cnn(h) #出口虚假图片
return Model(input=[latent, image_class], output=fake_image)
概念判别模型,输入(1,28,28)图片,输出两独价值,一个凡甄别模型认为这张图纸是否是虚假图片,另一个凡识别模型认为就第图片所属分类。

def build_generator(latent_size):
cnn = Sequential()
cnn.add(Dense(1024, input_dim=latent_size, activation=’relu’))
cnn.add(Dense(128 * 7 * 7, activation=’relu’))
cnn.add(Reshape((128, 7, 7)))
#上采样,图你尺寸变为 14X14
cnn.add(UpSampling2D(size=(2,2)))
cnn.add(Convolution2D(256, 5, 5, border_mode=’same’, activation=’relu’,
init=’glorot_normal’))
#齐采样,图像尺寸变为28X28
cnn.add(UpSampling2D(size=(2,2)))
cnn.add(Convolution2D(128, 5, 5, border_mode=’same’, activation=’relu’,
init=’glorot_normal’))
#规约到1个通道
cnn.add(Convolution2D(1, 2, 2, border_mode=’same’, activation=’tanh’,
init=’glorot_normal’))
#浮动模型输入层,特征向量
latent = Input(shape=(latent_size, ))
#浮动模型输入层,标记
image_class = Input(shape=(1,), dtype=’int32′)
cls = Flatten()(Embedding(10, latent_size,
init=’glorot_normal’)(image_class))
h = merge([latent, cls], mode=’mul’)
fake_image = cnn(h) #出口虚假图片
return Model(input=[latent, image_class], output=fake_image)
概念判别模型,输入(1,28,28)图片,输出两个价值,一个凡是辨模型认为就张图片是否是假图片,另一个是可辨模型认为当下第图片所属分类。

def build_discriminator();
#运激活函数Leaky ReLU来替换标准的卷积神经网络中的激活函数
cnn = Wequential()
cnn.add(Convolution2D(32, 3, 3, border_mode=’same’, subsample=(2, 2),
input_shape=(1, 28, 28)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(64, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(128, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(256, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Flatten())
image = Input(shape=(1, 28, 28))
features = cnn(image)
#产生些许个出口
#输出真假值,范围在0~1
fake = Dense(1, activation=’sigmoid’,name=’generation’)(features)
#赞助分类器,输出图片分类
aux = Dense(10, activation=’softmax’, name=’auxiliary’)(features)
return Model(input=image, output=[fake, aux])
教练过程,50车轮(epoch),把权重保存,每轮把假冒伪劣数据生成图处保存,观察虚假数据演化过程。

def build_discriminator();
#采用激活函数Leaky ReLU来替换标准的卷积神经网络中的激活函数
cnn = Wequential()
cnn.add(Convolution2D(32, 3, 3, border_mode=’same’, subsample=(2, 2),
input_shape=(1, 28, 28)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(64, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(128, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(256, 3, 3, border_mode=’same’, subsample=(1,
1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Flatten())
image = Input(shape=(1, 28, 28))
features = cnn(image)
#发出半点只出口
#输出真假值,范围在0~1
fake = Dense(1, activation=’sigmoid’,name=’generation’)(features)
#帮分类器,输出图片分类
aux = Dense(10, activation=’softmax’, name=’auxiliary’)(features)
return Model(input=image, output=[fake, aux])
教练过程,50轮(epoch),把权重保存,每轮把假冒伪劣数据生成图处保存,观察虚假数据演化过程。

if __name__ ==’__main__’:
#概念超参数
nb_epochs = 50
batch_size = 100
latent_size = 100
#优化器学习率
adam_lr = 0.0002
adam_beta_l = 0.5
#构建判别网络
discriminator = build_discriminator()
discriminator.compile(optimizer=adam(lr=adam_lr,
beta_l=adam_beta_l), loss=’binary_crossentropy’)
latent = Input(shape=(lastent_size, ))
image_class = Input(shape-(1, ), dtype=’int32′)
#变动组合型
discriminator.trainable = False
fake, aux = discriminator(fake)
combined = Model(input=[latent, image_class], output=[fake, aux])
combined.compile(optimizer=Adam(lr=adam_lr, beta_l=adam_beta_1),
loss=[‘binary_crossentropy’, ‘sparse_categorical_crossentropy’])
#以mnist数据转发为(…,1,28,28)维度,取值范围也[-1,1]
(X_train,y_train),(X_test,y_test) = mnist.load_data()
X_train = (X_train.astype(np.float32) – 127.5) / 127.5
X_train = np.expand_dims(X_train, axis=1)
X_test = (X_test.astype(np.float32) – 127.5) / 127.5
X_test = np.expand_dims(X_test, axis=1)
num_train, num_test = X_train.shape[0], X_test.shape[0]
train_history = defaultdict(list)
test_history = defaultdict(list)
for epoch in range(epochs):
print(‘Epoch {} of {}’.format(epoch + 1, epochs))
num_batches = int(X_train.shape[0] / batch_size)
progress_bar = Progbar(target=num_batches)
epoch_gen_loss = []
epoch_disc_loss = []
for index in range(num_batches):
progress_bar.update(index)
#来一个批次的噪音数据
noise = np.random.uniform(-1, 1, (batch_size, latent_size))
# 获取一个批次底真人真事数据
image_batch = X_train[index * batch_size:(index + 1) *
batch_size]
label_batch = y_train[index * batch_size:(index + 1) *
batch_size]
# 生成有噪声标记
sampled_labels = np.random.randint(0, 10, batch_size)
# 产生一个批次之虚图片
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=0)
X = np.concatenate((image_batch, generated_images))
y = np.array([1] * batch_size + [0] * batch_size)
aux_y = np.concatenate((label_batch, sampled_labels), axis=0)
epoch_disc_loss.append(discriminator.train_on_batch(X, [y,
aux_y]))
# 产生两单批次噪声和符号
noise = np.random.uniform(-1, 1, (2 * batch_size, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * batch_size)
# 训练转变模型来掩人耳目判别模型,输出真/假都使为真
trick = np.ones(2 * batch_size)
epoch_gen_loss.append(combined.train_on_batch(
[noise, sampled_labels.reshape((-1, 1))],
[trick, sampled_labels]))
print(‘\nTesting for epoch {}:’.format(epoch + 1))
# 评估测试集,产生一个初批次噪声数据
noise = np.random.uniform(-1, 1, (num_test, latent_size))
sampled_labels = np.random.randint(0, 10, num_test)
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=False)
X = np.concatenate((X_test, generated_images))
y = np.array([1] * num_test + [0] * num_test)
aux_y = np.concatenate((y_test, sampled_labels), axis=0)
# 判别模型是否能够鉴别
discriminator_test_loss = discriminator.evaluate(
X, [y, aux_y], verbose=False)
discriminator_train_loss = np.mean(np.array(epoch_disc_loss),
axis=0)
# 创建两个批次新噪声数据
noise = np.random.uniform(-1, 1, (2 * num_test, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * num_test)
trick = np.ones(2 * num_test)
generator_test_loss = combined.evaluate(
[noise, sampled_labels.reshape((-1, 1))],
[trick, sampled_labels], verbose=False)
generator_train_loss = np.mean(np.array(epoch_gen_loss), axis=0)
# 损失值等性能指标记录下来,并出口
train_history[‘generator’].append(generator_train_loss)
train_history[‘discriminator’].append(discriminator_train_loss)
test_history[‘generator’].append(generator_test_loss)
test_history[‘discriminator’].append(discriminator_test_loss)
print(‘{0:<22s} | {1:4s} | {2:15s} | {3:5s}’.format(
‘component’, *discriminator.metrics_names))
print(‘-‘ * 65)
ROW_FMT = ‘{0:<22s} | {1:<4.2f} | {2:<15.2f} | {3:<5.2f}’
print(ROW_FMT.format(‘generator (train)’,
*train_history[‘generator’][-1]))
print(ROW_FMT.format(‘generator (test)’,
*test_history[‘generator’][-1]))
print(ROW_FMT.format(‘discriminator (train)’,
*train_history[‘discriminator’][-1]))
print(ROW_FMT.format(‘discriminator (test)’,
*test_history[‘discriminator’][-1]))
# 每个epoch保存一不好权重
generator.save_weights(
‘params_generator_epoch_{0:03d}.hdf5’.format(epoch), True)
discriminator.save_weights(
‘params_discriminator_epoch_{0:03d}.hdf5’.format(epoch), True)
# 生成有可视化虚假数字看演化过程
noise = np.random.uniform(-1, 1, (100, latent_size))
sampled_labels = np.array([
[i] * 10 for i in range(10)
]).reshape(-1, 1)
generated_images = generator.predict(
[noise, sampled_labels], verbose=0)
# 整理及一个方格
img = (np.concatenate([r.reshape(-1, 28)
for r in np.split(generated_images, 10)
], axis=-1) * 127.5 + 127.5).astype(np.uint8)
Image.fromarray(img).save(
‘plot_epoch_{0:03d}_generated.png’.format(epoch))
pickle.dump({‘train’: train_history, ‘test’: test_history},
open(‘acgan-history.pkl’, ‘wb’))

if __name__ ==’__main__’:
#概念超参数
nb_epochs = 50
batch_size = 100
latent_size = 100
#优化器学习率
adam_lr = 0.0002
adam_beta_l = 0.5
#构建判别网络
discriminator = build_discriminator()
discriminator.compile(optimizer=adam(lr=adam_lr,
beta_l=adam_beta_l), loss=’binary_crossentropy’)
latent = Input(shape=(lastent_size, ))
image_class = Input(shape-(1, ), dtype=’int32′)
#变迁组合型
discriminator.trainable = False
fake, aux = discriminator(fake)
combined = Model(input=[latent, image_class], output=[fake, aux])
combined.compile(optimizer=Adam(lr=adam_lr, beta_l=adam_beta_1),
loss=[‘binary_crossentropy’, ‘sparse_categorical_crossentropy’])
#以mnist数据转发为(…,1,28,28)维度,取值范围吗[-1,1]
(X_train,y_train),(X_test,y_test) = mnist.load_data()
X_train = (X_train.astype(np.float32) – 127.5) / 127.5
X_train = np.expand_dims(X_train, axis=1)
X_test = (X_test.astype(np.float32) – 127.5) / 127.5
X_test = np.expand_dims(X_test, axis=1)
num_train, num_test = X_train.shape[0], X_test.shape[0]
train_history = defaultdict(list)
test_history = defaultdict(list)
for epoch in range(epochs):
print(‘Epoch {} of {}’.format(epoch + 1, epochs))
num_batches = int(X_train.shape[0] / batch_size)
progress_bar = Progbar(target=num_batches)
epoch_gen_loss = []
epoch_disc_loss = []
for index in range(num_batches):
progress_bar.update(index)
#起一个批次之噪声数据
noise = np.random.uniform(-1, 1, (batch_size, latent_size))
# 获取一个批次的实数据
image_batch = X_train[index * batch_size:(index + 1) *
batch_size]
label_batch = y_train[index * batch_size:(index + 1) *
batch_size]
# 生成有噪音标记
sampled_labels = np.random.randint(0, 10, batch_size)
# 产生一个批次底仿真图片
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=0)
X = np.concatenate((image_batch, generated_images))
y = np.array([1] * batch_size + [0] * batch_size)
aux_y = np.concatenate((label_batch, sampled_labels), axis=0)
epoch_disc_loss.append(discriminator.train_on_batch(X, [y,
aux_y]))
# 产生两单批次噪声和标记
noise = np.random.uniform(-1, 1, (2 * batch_size, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * batch_size)
# 训练转变模型来欺骗判别模型,输出真/假都如为实在
trick = np.ones(2 * batch_size)
epoch_gen_loss.append(combined.train_on_batch(
[noise, sampled_labels.reshape((-1, 1))],
[trick, sampled_labels]))
print(‘\nTesting for epoch {}:’.format(epoch + 1))
# 评估测试集,产生一个初批次噪声数据
noise = np.random.uniform(-1, 1, (num_test, latent_size))
sampled_labels = np.random.randint(0, 10, num_test)
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=False)
X = np.concatenate((X_test, generated_images))
y = np.array([1] * num_test + [0] * num_test)
aux_y = np.concatenate((y_test, sampled_labels), axis=0)
# 判别模型是否能鉴别
discriminator_test_loss = discriminator.evaluate(
X, [y, aux_y], verbose=False)
discriminator_train_loss = np.mean(np.array(epoch_disc_loss),
axis=0)
# 创建两单批次新噪声数据
noise = np.random.uniform(-1, 1, (2 * num_test, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * num_test)
trick = np.ones(2 * num_test)
generator_test_loss = combined.evaluate(
[noise, sampled_labels.reshape((-1, 1))],
[trick, sampled_labels], verbose=False)
generator_train_loss = np.mean(np.array(epoch_gen_loss), axis=0)
# 损失值等性能指标记录下来,并出口
train_history[‘generator’].append(generator_train_loss)
train_history[‘discriminator’].append(discriminator_train_loss)
test_history[‘generator’].append(generator_test_loss)
test_history[‘discriminator’].append(discriminator_test_loss)
print(‘{0:<22s} | {1:4s} | {2:15s} | {3:5s}’.format(
‘component’, *discriminator.metrics_names))
print(‘-‘ * 65)
ROW_FMT = ‘{0:<22s} | {1:<4.2f} | {2:<15.2f} | {3:<5.2f}’
print(ROW_FMT.format(‘generator (train)’,
*train_history[‘generator’][-1]))
print(ROW_FMT.format(‘generator (test)’,
*test_history[‘generator’][-1]))
print(ROW_FMT.format(‘discriminator (train)’,
*train_history[‘discriminator’][-1]))
print(ROW_FMT.format(‘discriminator (test)’,
*test_history[‘discriminator’][-1]))
# 每个epoch保存一涂鸦权重
generator.save_weights(
‘params_generator_epoch_{0:03d}.hdf5’.format(epoch), True)
discriminator.save_weights(
‘params_discriminator_epoch_{0:03d}.hdf5’.format(epoch), True)
# 生成有可视化虚假数字看演化过程
noise = np.random.uniform(-1, 1, (100, latent_size))
sampled_labels = np.array([
[i] * 10 for i in range(10)
]).reshape(-1, 1)
generated_images = generator.predict(
[noise, sampled_labels], verbose=0)
# 整理及一个方格
img = (np.concatenate([r.reshape(-1, 28)
for r in np.split(generated_images, 10)
], axis=-1) * 127.5 + 127.5).astype(np.uint8)
Image.fromarray(img).save(
‘plot_epoch_{0:03d}_generated.png’.format(epoch))
pickle.dump({‘train’: train_history, ‘test’: test_history},
open(‘acgan-history.pkl’, ‘wb’))

训练了,创建3类文件。params_discriminator_epoch_{{epoch_number}}.hdf5,判别模型权重参数。params_generator_epoch_{{epoch_number}}.hdf5,生成模型权重参数。plot_epoch_{{epoch_number}}_generated.png

教练了,创建3类文件。params_discriminator_epoch_{{epoch_number}}.hdf5,判别模型权重参数。params_generator_epoch_{{epoch_number}}.hdf5,生成模型权重参数。plot_epoch_{{epoch_number}}_generated.png

生成式对抗网络改进。生成式对抗网络(generative adversarial
network,GAN)在管监控上好实用。常规生成式对抗网络判别器使用Sigmoid交叉熵损失函数,学习过程梯度消失。Wasserstein生成式对抗网络(Wasserstein
generative adversarial
network,WGAN),使用Wasserstein距离度量,而休是Jensen-Shannon散度(Jensen-Shannon
divergence,JSD)。使用最小二随着生成式对抗网络(least squares generative
adversarial network,LSGAN),判别模型用极端小平方损失小函数(least squares
loss function)。Sebastian Nowozin、Botond Cseke、Ryota
Tomioka论文《f-GAN: Training Generative Neural Samplers using
Variational Divergence Minimization》。

生成式对抗网络改进。生成式对抗网络(generative adversarial
network,GAN)在无监控上不行实用。常规生成式对抗网络判别器使用Sigmoid交叉熵损失函数,学习过程梯度消失。Wasserstein生成式对抗网络(Wasserstein
generative adversarial
network,WGAN),使用Wasserstein距离度量,而不是Jensen-Shannon散度(Jensen-Shannon
divergence,JSD)。使用最小二乘机生成式对抗网络(least squares generative
adversarial network,LSGAN),判别模型用最好小平方损失小函数(least squares
loss function)。Sebastian Nowozin、Botond Cseke、Ryota
Tomioka论文《f-GAN: Training Generative Neural Samplers using
Variational Divergence Minimization》。

参考资料:
《TensorFlow技术解析及实战》

参考资料:
《TensorFlow技术解析和实战》

欢迎付费咨询(150长诸小时),我之微信:qingxingfengzi

迎接付费咨询(150正各个时),我之微信:qingxingfengzi

相关文章