Python深度学习（11）：GAN生成青蛙图片

算法简介

GAN最直观的解释就是博弈，同时训练两个网络（生成网络和判别网络），二者训练都是为了打败彼此。生成网络将随机潜在向量转换为图像，判别器试图分辨真实图像与生成图像。
生成网络：以一张随机向量作为输入，解码为合成图像
判别网络：输入图像，输出真或假的类别
实验中搭建的是DCGAN深度卷积生成式对抗网络，即生成网络和判别网络都是深度卷积网络，具体实现流程如下：
（1）潜在空间抽取随机噪声
（2）生成网络利用这些随机噪声生成图像
（3）将生成图像与真实图像打上标签，并混合，
（4）利用混合后的图像集去训练判别网络
（5）回到（1）

代码实现

import keras
from keras import layers
import numpy as np
import os
from keras.preprocessing import image

latent_dim = 32
height = 32
width = 32
channels = 3

generator_input = keras.Input(shape=(latent_dim,))

# 生成模型
# 将输入转换为16*16的128个通道的特征图
x = layers.Dense(128 * 16 * 16)(generator_input)
x = layers.LeakyReLU()(x)
x = layers.Reshape((16, 16, 128))(x)

x = layers.Conv2D(256, 5, padding='same')(x)
x = layers.LeakyReLU()(x)

x = layers.Conv2DTranspose(256, 4, strides=2, padding='same')(x)  # 使用Conv2DTranspose层对图像进行上采样
x = layers.LeakyReLU()(x)

x = layers.Conv2D(256, 5, padding='same')(x)
x = layers.LeakyReLU()(x)
x = layers.Conv2D(256, 5, padding='same')(x)
x = layers.LeakyReLU()(x)

x = layers.Conv2D(channels, 7, activation='tanh', padding='same')(x)
generator = keras.models.Model(generator_input, x)  # 将生成器实例化（由向量映射到图像）
print(generator.summary())

# 判别器模型
discriminator_input = layers.Input(shape=(height, width, channels))
x = layers.Conv2D(128, 3)(discriminator_input)
x = layers.LeakyReLU()(x)
x = layers.Conv2D(128, 4, strides=2)(x)
x = layers.LeakyReLU()(x)
x = layers.Conv2D(128, 4, strides=2)(x)
x = layers.LeakyReLU()(x)
x = layers.Flatten()(x)

x = layers.Dropout(0.4)(x)

x = layers.Dense(1, activation='sigmoid')(x)

discriminator = keras.models.Model(discriminator_input, x)  # 将判别器模型实例化
discriminator.summary()

discriminator_optimizer = keras.optimizers.RMSprop(lr=0.0008, clipvalue=1.0, decay=1e-8)
discriminator.compile(optimizer=discriminator_optimizer, loss='binary_crossentropy')

discriminator.trainable = False  # 将判别器权重设置为不可训练

gan_input = keras.Input(shape=(latent_dim,))
gan_output = discriminator(generator(gan_input))
gan = keras.models.Model(gan_input, gan_output)  # gan模型实例化

gan_optimizer = keras.optimizers.RMSprop(lr=0.0004, clipvalue=1.0, decay=1e-8)
gan.compile(optimizer=gan_optimizer, loss='binary_crossentropy')

(x_train, y_train), (_, _) = keras.datasets.cifar10.load_data()  # 加载cifar10训练集
x_train = x_train[y_train.flatten() == 6]  # 选择其中的青蛙图像
x_train = x_train.reshape((x_train.shape[0],) + (height, width, channels)).astype('float32') / 255.  # 数据标准化

iterations = 10000
batch_size = 20
save_dir = './gan_png'  # 指定保存生成图像的目录

start = 0
for step in range(iterations):
    random_latent_vectors = np.random.normal(size=(batch_size, latent_dim))  # 在潜在空间采样随机点
    generated_images = generator.predict(random_latent_vectors)  # 将随机点解码为图像

    stop = start + batch_size
    real_images = x_train[start: stop]
    combined_images = np.concatenate([generated_images, real_images])  # 将生成的图像与真实图像混合

    labels = np.concatenate([np.ones((batch_size, 1)), np.zeros((batch_size, 1))])  # 所有图像的标签

    labels += 0.05 * np.random.random(labels.shape)  # 向标签中添加随机噪声

    d_loss = discriminator.train_on_batch(combined_images, labels)  # 训练判别器

    random_latent_vectors = np.random.normal(size=(batch_size, latent_dim))  # 在潜在空间采样随机点

    misleading_targets = np.zeros((batch_size, 1))  # 合并标签

    a_loss = gan.train_on_batch(random_latent_vectors, misleading_targets)  # 冻结判别器，训练生成器

    # 保存与展示
    start += batch_size
    if start > len(x_train) - batch_size:
        start = 0
    if step % 100 == 0:
        gan.save_weights('gan.h5')

        print('discriminator loss:', d_loss)
        print('adversarial loss:', a_loss)

        img = image.array_to_img(generated_images[0] * 255., scale=False)
        img.save(os.path.join(save_dir, 'generated_frog' + str(step) + '.png'))

        img = image.array_to_img(real_images[0] * 255., scale=False)
        img.save(os.path.join(save_dir, 'real_frog' + str(step) + '.png'))

运行结果：