上传空间站的注意事项,我们为什么选择做电子商务网站,吉林市网页设计,东莞it外包前言
生成对抗网络#xff08;Generative Adversarial Networks#xff0c;简称GANs#xff09;是一种用于生成新样本的机器学习模型。它由两个主要组件组成#xff1a;生成器#xff08;Generator#xff09;和判别器#xff08;Discriminator#xff09;。生成器尝试…前言
生成对抗网络Generative Adversarial Networks简称GANs是一种用于生成新样本的机器学习模型。它由两个主要组件组成生成器Generator和判别器Discriminator。生成器尝试生成与训练数据相似的新样本而判别器则试图区分生成器生成的样本和真实训练数据。 下面是一个简单的对抗生成网络的入门例子用于生成手写数字图像
实现过程
1、导入必要的库和模块
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten, Reshape
from tensorflow.keras.layers import Conv2D, Conv2DTranspose
from tensorflow.keras.optimizers import Adam2、加载MNIST数据集
(x_train, _), (_, _) mnist.load_data()
x_train x_train / 255.0
x_train np.expand_dims(x_train, axis3)3、定义生成器模型
generator Sequential()
generator.add(Dense(7*7*128, input_shape(100,), activationrelu))
generator.add(Reshape((7, 7, 128)))
generator.add(Conv2DTranspose(64, (3, 3), strides(2, 2), paddingsame, activationrelu))
generator.add(Conv2DTranspose(1, (3, 3), strides(2, 2), paddingsame, activationsigmoid))4、定义判别器模型
discriminator Sequential()
discriminator.add(Conv2D(64, (3, 3), strides(2, 2), paddingsame, input_shape(28, 28, 1), activationrelu))
discriminator.add(Conv2D(128, (3, 3), strides(2, 2), paddingsame, activationrelu))
discriminator.add(Flatten())
discriminator.add(Dense(1, activationsigmoid))
5、编译判别器模型
discriminator.compile(lossbinary_crossentropy, optimizerAdam(learning_rate0.0002, beta_10.5), metrics[accuracy])
6、冻结判别器模型的权重
discriminator.trainable False
7、定义GAN模型
gan Sequential()
gan.add(generator)
gan.add(discriminator)
8、编译GAN模型
gan.compile(lossbinary_crossentropy, optimizerAdam(learning_rate0.0002, beta_10.5))
9、定义训练函数
def train_gan(epochs, batch_size, sample_interval):for epoch in range(epochs):# 生成随机噪声作为输入noise np.random.normal(0, 1, (batch_size, 100))# 生成假样本generated_images generator.predict(noise)# 从真实样本中随机选择一批样本real_images x_train[np.random.randint(0, x_train.shape[0], batch_size)]# 训练判别器discriminator_loss_real discriminator.train_on_batch(real_images, np.ones((batch_size, 1)))discriminator_loss_fake discriminator.train_on_batch(generated_images, np.zeros((batch_size, 1)))discriminator_loss 0.5 * np.add(discriminator_loss_real, discriminator_loss_fake)# 训练生成器noise np.random.normal(0, 1, (batch_size, 100))generator_loss gan.train_on_batch(noise, np.ones((batch_size, 1)))# 打印损失if epoch % sample_interval 0:print(fEpoch {epoch}/{epochs}, Discriminator Loss: {discriminator_loss[0]}, Generator Loss: {generator_loss})# 保存生成的图像save_images(epoch)
10、保存生成的图像
def save_images(epoch):rows, cols 5, 5noise np.random.normal(0, 1, (rows * cols, 100))generated_images generator.predict(noise)generated_images 0.5 * generated_images 0.5fig, axs plt.subplots(rows, cols)idx 0for i in range(rows):for j in range(cols):axs[i, j].imshow(generated_images[idx, :, :, 0], cmapgray)axs[i, j].axis(off)idx 1fig.savefig(fgan_images/mnist_{epoch}.png)plt.close()
11、训练GAN模型
epochs 10000
batch_size 128
sample_interval 1000 完整代码
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten, Reshape
from tensorflow.keras.layers import Conv2D, Conv2DTranspose
from tensorflow.keras.optimizers import Adam# 加载MNIST数据集
(x_train, _), (_, _) mnist.load_data()
x_train x_train / 255.0
x_train np.expand_dims(x_train, axis3)# 定义生成器模型
generator Sequential()
generator.add(Dense(7*7*128, input_shape(100,), activationrelu))
generator.add(Reshape((7, 7, 128)))
generator.add(Conv2DTranspose(64, (3, 3), strides(2, 2), paddingsame, activationrelu))
generator.add(Conv2DTranspose(1, (3, 3), strides(2, 2), paddingsame, activationsigmoid))# 定义判别器模型
discriminator Sequential()
discriminator.add(Conv2D(64, (3, 3), strides(2, 2), paddingsame, input_shape(28, 28, 1), activationrelu))
discriminator.add(Conv2D(128, (3, 3), strides(2, 2), paddingsame, activationrelu))
discriminator.add(Flatten())
discriminator.add(Dense(1, activationsigmoid))# 编译判别器模型
discriminator.compile(lossbinary_crossentropy, optimizerAdam(learning_rate0.0002, beta_10.5), metrics[accuracy])# 冻结判别器模型的权重
discriminator.trainable False# 定义GAN模型
gan Sequential()
gan.add(generator)
gan.add(discriminator)# 编译GAN模型
gan.compile(lossbinary_crossentropy, optimizerAdam(learning_rate0.0002, beta_10.5))# 定义训练函数
def train_gan(epochs, batch_size, sample_interval):for epoch in range(epochs):# 生成随机噪声作为输入noise np.random.normal(0, 1, (batch_size, 100))# 生成假样本generated_images generator.predict(noise)# 从真实样本中随机选择一批样本real_images x_train[np.random.randint(0, x_train.shape[0], batch_size)]# 训练判别器discriminator_loss_real discriminator.train_on_batch(real_images, np.ones((batch_size, 1)))discriminator_loss_fake discriminator.train_on_batch(generated_images, np.zeros((batch_size, 1)))discriminator_loss 0.5 * np.add(discriminator_loss_real, discriminator_loss_fake)# 训练生成器noise np.random.normal(0, 1, (batch_size, 100))generator_loss gan.train_on_batch(noise, np.ones((batch_size, 1)))# 打印损失if epoch % sample_interval 0:print(fEpoch {epoch}/{epochs}, Discriminator Loss: {discriminator_loss[0]}, Generator Loss: {generator_loss})# 保存生成的图像save_images(epoch)# 保存生成的图像
def save_images(epoch):rows, cols 5, 5noise np.random.normal(0, 1, (rows * cols, 100))generated_images generator.predict(noise)generated_images 0.5 * generated_images 0.5fig, axs plt.subplots(rows, cols)idx 0for i in range(rows):for j in range(cols):axs[i, j].imshow(generated_images[idx, :, :, 0], cmapgray)axs[i, j].axis(off)idx 1fig.savefig(fgan_images/mnist_{epoch}.png)plt.close()# 训练GAN模型
epochs 10000
batch_size 128
sample_interval 1000train_gan(epochs, batch_size, sample_interval)训练结果 这个例子使用了MNIST数据集生成手写数字图像。生成器和判别器模型使用了卷积神经网络的结构。在训练过程中生成器试图生成逼真的手写数字图像而判别器则试图区分真实图像和生成图像。通过反复迭代训练生成器和判别器GAN模型能够逐渐生成更逼真的手写数字图像。生成的图像会保存在gan_images文件夹中。
