AI 图像生成入门：原理、工具和实践 | 极客日志

AI 图像生成入门：原理、工具和实践 | 极客日志

from keras.models import Sequential
from keras.layers import Dense, Reshape, Flatten, Conv2D, Conv2DTranspose
from keras.optimizers import Adam

# 生成器定义
generator = Sequential()
generator.add(Dense(128*7*7, activation='relu', input_dim=100))
generator.add(Reshape((7, 7, 128)))
generator.add(Conv2DTranspose(64, kernel_size=3, strides=2, activation='relu'))
generator.add(Conv2DTranspose(1, kernel_size=3, strides=2, activation='tanh'))

# 判别器定义
discriminator = Sequential()
discriminator.add(Conv2D(64, kernel_size=3, strides=2, input_shape=(28, 28, 1)))
discriminator.add(Flatten())
discriminator.add(Dense(1, activation='sigmoid'))

# 判别器编译
discriminator.compile(loss='binary_crossentropy', optimizer=Adam(), metrics=['accuracy'])

# 组合模型
gan = Sequential([generator, discriminator])
gan.compile(loss='binary_crossentropy', optimizer=Adam())

# 训练过程需要加载 MNIST 数据集，然后对生成器和判别器进行交替训练，逐步提升图像生成质量。

import torch
import torch.nn as nn
import torch.optim as optim

class VAE(nn.Module):
    def __init__(self):
        super(VAE, self).__init__()
        # 编码器
        self.encoder = nn.Sequential(
            nn.Linear(784, 400),
            nn.ReLU(),
            nn.Linear(400, 20*2) # 输出均值和方差
        )
        # 解码器
        self.decoder = nn.Sequential(
            nn.Linear(20, 400),
            nn.ReLU(),
            nn.Linear(400, 784),
            nn.Sigmoid()
        )

    def reparameterize(self, mu, logvar):
        std = torch.exp(0.5*logvar)
        eps = torch.randn_like(std)
        return mu + eps*std

    def forward(self, x):
        h = self.encoder(x.view(-1, 784))
        mu, logvar = h.chunk(2, dim=1)
        z = self.reparameterize(mu, logvar)
        return self.decoder(z), mu, logvar

# 初始化和训练
model = VAE()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
# 训练循环中需要计算重构损失和 KL 散度损失，通过优化器更新参数，使模型学会生成与原始手写数字相似的图像。

import tensorflow as tf
import tensorflow_hub as hub

# 加载预训练的图像到图像生成模型（如 cyclegan）
model = hub.load('https://tfhub.dev/tensorflow/cyclegan/1')

# 假设 input_sketch 是一张草图图像（预处理后的张量形式，如归一化到 [0,1] 范围）
# 将草图转换为彩色图像
generated_image = model(input_sketch, training=False)

# 可视化生成的图像
import matplotlib.pyplot as plt
plt.imshow(generated_image[0])
plt.show()