sgan and cgan

SGAN和CGAN都是生成对抗网络（GAN）的变体，用于生成逼真的图像。

SGAN是一种条件生成对抗网络，它的生成器和判别器都是基于条件的。这意味着生成器和判别器都会接收一个额外的条件作为输入，以帮助生成更加准确的图像。在SGAN中，条件可能是类别标签，文本描述或其他信息。

CGAN是一种带有类别标签的生成对抗网络。它的生成器和判别器都接收一个类别标签作为额外的输入。这样，生成器可以生成与特定类别相关的图像，而判别器则可以识别图像是否与给定的类别相关。

总的来说，SGAN和CGAN都是用于生成高质量图像的强大技术，它们在不同的应用场景中都有广泛的应用。

相关问题

CGAN代码

以下是一个简单的CGAN的Python代码示例，用于生成手写数字图像。假设我们想要在MNIST数据集上生成数字4的图像：

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Dense, Flatten, Reshape, Conv2DTranspose, Conv2D, LeakyReLU, Dropout, Input, Embedding, Concatenate
from tensorflow.keras import Model

# 加载MNIST数据集
(x_train, y_train), (_, _) = tf.keras.datasets.mnist.load_data()
x_train = x_train[y_train == 4]  # 选择数字4的图像
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1).astype('float32')
x_train = (x_train - 127.5) / 127.5  # 将像素值缩放到[-1, 1]之间

BUFFER_SIZE = 60000
BATCH_SIZE = 256
EPOCHS = 50
NUM_CLS = 10  # 数字的类别数
NOISE_DIM = 100

# 用于生成假图像的生成器网络
def make_generator_model():
    model = tf.keras.Sequential()
    model.add(Dense(7*7*256, use_bias=False, input_shape=(NOISE_DIM+NUM_CLS,)))
    model.add(BatchNormalization())
    model.add(LeakyReLU())

    model.add(Reshape((7, 7, 256)))
    assert model.output_shape == (None, 7, 7, 256)  # 注意：batch size 没有限制
    model.add(Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
    assert model.output_shape == (None, 7, 7, 128)
    model.add(BatchNormalization())
    model.add(LeakyReLU())

    model.add(Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
    assert model.output_shape == (None, 14, 14, 64)
    model.add(BatchNormalization())
    model.add(LeakyReLU())

    model.add(Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
    assert model.output_shape == (None, 28, 28, 1)

    return model

# 用于判别真假图像的判别器网络
def make_discriminator_model():
    model = tf.keras.Sequential()
    model.add(Conv2D(64, (5, 5), strides=(2, 2), padding='same',
                                     input_shape=[28, 28, 1]))
    model.add(LeakyReLU())
    model.add(Dropout(0.3))

    model.add(Conv2D(128, (5, 5), strides=(2, 2), padding='same'))
    model.add(LeakyReLU())
    model.add(Dropout(0.3))

    model.add(Flatten())
    model.add(Dense(1))

    return model

# 将类别信息转换成one-hot编码
def make_one_hot(labels):
    one_hot = tf.one_hot(labels, NUM_CLS)
    return tf.reshape(one_hot, (len(labels), NUM_CLS))

# 定义损失函数
cross_entropy = tf.keras.losses.BinaryCrossentropy(from_logits=True)
categorical_crossentropy = tf.keras.losses.CategoricalCrossentropy(from_logits=True)

# 定义生成器的损失函数
def generator_loss(fake_output, labels):
    return cross_entropy(tf.ones_like(fake_output), fake_output) + categorical_crossentropy(labels, fake_output)

# 定义判别器的损失函数
def discriminator_loss(real_output, fake_output):
    real_loss = cross_entropy(tf.ones_like(real_output), real_output)
    fake_loss = cross_entropy(tf.zeros_like(fake_output), fake_output)
    total_loss = real_loss + fake_loss
    return total_loss

# 定义优化器
generator_optimizer = tf.keras.optimizers.Adam(1e-4)
discriminator_optimizer = tf.keras.optimizers.Adam(1e-4)

# 定义生成器和判别器
generator = make_generator_model()
discriminator = make_discriminator_model()

# 定义训练步骤
@tf.function
def train_step(images, labels):
    noise = tf.random.normal([BATCH_SIZE, NOISE_DIM])
    gen_input = tf.concat([noise, labels], axis=1)

    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:
        generated_images = generator(gen_input, training=True)

        real_output = discriminator(images, training=True)
        fake_output = discriminator(generated_images, training=True)

        gen_loss = generator_loss(fake_output, labels)
        disc_loss = discriminator_loss(real_output, fake_output)

    gradients_of_generator = gen_tape.gradient(gen_loss, generator.trainable_variables)
    gradients_of_discriminator = disc_tape.gradient(disc_loss, discriminator.trainable_variables)

    generator_optimizer.apply_gradients(zip(gradients_of_generator, generator.trainable_variables))
    discriminator_optimizer.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))

# 开始训练
for epoch in range(EPOCHS):
    for i in range(x_train.shape[0] // BATCH_SIZE):
        images = x_train[i * BATCH_SIZE:(i + 1) * BATCH_SIZE]
        labels = make_one_hot(np.ones(BATCH_SIZE) * 4)  # 生成数字4的图像
        train_step(images, labels)

    # 每个epoch结束后生成一张图像
    noise = tf.random.normal([1, NOISE_DIM])
    label = make_one_hot(np.array([4]))
    gen_input = tf.concat([noise, label], axis=1)
    generated_image = generator(gen_input, training=False)
    generated_image = generated_image * 127.5 + 127.5  # 将像素值还原到[0, 255]之间
    generated_image = generated_image.numpy().squeeze().astype('uint8')
    plt.imshow(generated_image, cmap='gray')
    plt.axis('off')
    plt.show()

这段代码中使用了MNIST数据集，通过生成器生成数字4的图像。在训练过程中，我们将生成器和判别器进行交替训练，并通过优化器更新网络参数。

cgan代码

以下是一个简单的CGAN的代码示例：

import torch
import torch.nn as nn
import numpy as np

# 定义 generator 和 discriminator 的网络结构
class Generator(nn.Module):
    def __init__(self, input_dim, output_dim, hidden_dim):
        super(Generator, self).__init__()
        self.fc1 = nn.Linear(input_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, hidden_dim)
        self.fc3 = nn.Linear(hidden_dim, output_dim)

    def forward(self, x, y):
        z = torch.cat([x, y], 1)
        z = nn.functional.relu(self.fc1(z))
        z = nn.functional.relu(self.fc2(z))
        z = torch.tanh(self.fc3(z))
        return z

class Discriminator(nn.Module):
    def __init__(self, input_dim, hidden_dim):
        super(Discriminator, self).__init__()
        self.fc1 = nn.Linear(input_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, hidden_dim)
        self.fc3 = nn.Linear(hidden_dim, 1)

    def forward(self, x, y):
        z = torch.cat([x, y], 1)
        z = nn.functional.relu(self.fc1(z))
        z = nn.functional.relu(self.fc2(z))
        z = torch.sigmoid(self.fc3(z))
        return z

# 定义损失函数和优化器
criterion = nn.BCELoss() # 二分类交叉熵损失函数
G_optimizer = torch.optim.Adam(generator.parameters(), lr=0.0002, betas=(0.5, 0.999))
D_optimizer = torch.optim.Adam(discriminator.parameters(), lr=0.0002, betas=(0.5, 0.999))

# 定义训练函数
def train_GAN(num_epochs, data_loader):
    for epoch in range(num_epochs):
        for i, (real_data, real_label) in enumerate(data_loader):
            # 训练 discriminator
            D_optimizer.zero_grad()
            fake_label = torch.zeros(real_label.shape[0], 1)
            real_label = real_label.float().view(-1, 1)
            real_data = real_data.view(-1, input_dim)
            real_decision = discriminator(real_data, real_label)
            D_real_loss = criterion(real_decision, real_label)
            fake_data = generator(torch.randn(real_data.shape[0], z_dim), real_label)
            fake_decision = discriminator(fake_data, fake_label)
            D_fake_loss = criterion(fake_decision, fake_label)
            D_loss = D_real_loss + D_fake_loss
            D_loss.backward()
            D_optimizer.step()

            # 训练 generator
            G_optimizer.zero_grad()
            fake_label = torch.ones(real_label.shape[0], 1)
            fake_data = generator(torch.randn(real_data.shape[0], z_dim), real_label)
            fake_decision = discriminator(fake_data, fake_label)
            G_loss = criterion(fake_decision, fake_label)
            G_loss.backward()
            G_optimizer.step()

            # 打印训练信息
            if (i+1) % 20 == 0:
                print("Epoch [{}/{}], Step [{}/{}], Discriminator Loss: {:.4f}, Generator Loss: {:.4f}"
                      .format(epoch+1, num_epochs, i+1, len(data_loader), D_loss.item(), G_loss.item()))

# 定义数据集和参数
input_dim = 2 # 输入数据维度
output_dim = 2 # 输出数据维度
hidden_dim = 128 # 隐藏层维度
z_dim = 10 # 随机噪声维度
batch_size = 64 # 每批次训练数据量

# 准备数据集
data = np.random.multivariate_normal([0,0], [[1,0],[0,1]], size=1000) # 生成1000个二维高斯分布的数据
label = np.zeros((1000, 1)) # 标签全为0，表示数据是真实数据
dataset = torch.utils.data.TensorDataset(torch.tensor(data), torch.tensor(label))
data_loader = torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=True)

# 初始化 generator 和 discriminator
generator = Generator(input_dim+z_dim, output_dim, hidden_dim)
discriminator = Discriminator(input_dim+1, hidden_dim)

# 训练 GAN
num_epochs = 50
train_GAN(num_epochs, data_loader)

这个CGAN的代码中，Generator和Discriminator的网络结构都比较简单，只有3层全连接层。在训练过程中，我们先训练Discriminator，然后再训练Generator，交替进行，期望通过这个过程让Generator生成的假数据越来越逼近真实数据的分布。