变分自编码AVE器生成图像(Pytorch)

导入必要的包

import os
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
from torchvision import transforms
from torchvision.utils import save_image

import matplotlib.pyplot as plt
import matplotlib.image as mpimg

# 设备配置
# torch.cuda.set_device(1)    
# 这句用来设置pytorch在哪块GPU上运行，这里假设使用序号为1的这块GPU.
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# 在当前目录，创建不存在的目录ave_samples
sample_dir = 'ave_samples'
if not os.path.exists(sample_dir):
    os.makedirs(sample_dir)

定义一些超参数

image_size = 784
h_dim = 400
z_dim = 20
num_epochs = 30
batch_size = 128
learning_rate = 0.001

下载MNIST训练集

#这里因已下载，故download=False
dataset = torchvision.datasets.MNIST(root='data',
                                     train=True,
                                     transform=transforms.ToTensor(),
                                     download=False)

#数据加载
data_loader = torch.utils.data.DataLoader(dataset=dataset,
                                          batch_size=batch_size,
                                          shuffle=True)

定义AVE模型

class VAE(nn.Module):
    def __init__(self, image_size=784, h_dim=400, z_dim=20):
        super(VAE, self).__init__()
        self.fc1 = nn.Linear(image_size, h_dim)
        self.fc2 = nn.Linear(h_dim, z_dim)
        self.fc3 = nn.Linear(h_dim, z_dim)
        self.fc4 = nn.Linear(z_dim, h_dim)
        self.fc5 = nn.Linear(h_dim, image_size)

    def encode(self, x):
        h = F.relu(self.fc1(x))
        return self.fc2(h), self.fc3(h)

    # 用mu，log_var生成一个潜在空间点z，mu，log_var为两个统计参数，并假设这个假设分布能生成图像
    def reparameterize(self, mu, log_var):
        std = torch.exp(log_var / 2)
        eps = torch.randn_like(std) # std为随机采样的
        return mu + eps * std

    def decode(self, z):
        h = F.relu(self.fc4(z))
        return F.sigmoid(self.fc5(h))

    def forward(self, x):
        mu, log_var = self.encode(x)
        z = self.reparameterize(mu, log_var)
        x_reconst = self.decode(z)
        return x_reconst, mu, log_var


model = VAE().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)

开始训练模型

for epoch in range(num_epochs):
    model.train()
    for i, (x, _) in enumerate(data_loader):
        # 前向传播
        model.zero_grad()
        x = x.to(device).view(-1, image_size)
        x_reconst, mu, log_var = model(x)

        # Compute reconstruction loss and kl divergence
        # For KL divergence, see Appendix B in VAE paper or http://yunjey47.tistory.com/43
        reconst_loss = F.binary_cross_entropy(x_reconst, x, size_average=False)
        kl_div = - 0.5 * torch.sum(1 + log_var - mu.pow(2) - log_var.exp())

        # 反向传播及优化器
        loss = reconst_loss + kl_div    # 两者相加得总损失
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if (i + 1) % 100 == 0:
            print("Epoch[{}/{}], Step [{}/{}], Reconst Loss: {:.4f}, KL Div: {:.4f}"
                  .format(epoch + 1, num_epochs, i + 1, len(data_loader), reconst_loss.item(), kl_div.item()))

    with torch.no_grad():
        # 保存采样图像，即潜在向量Z通过解码器生成的新图像
        z = torch.randn(batch_size, z_dim).to(device)
        out = model.decode(z).view(-1, 1, 28, 28)
        save_image(out, os.path.join(sample_dir, 'sampled-{}.png'.format(epoch + 1)))

        # 保存重构图像，即原图像通过解码器生成的图像
        out, _, _ = model(x)
        x_concat = torch.cat([x.view(-1, 1, 28, 28), out.view(-1, 1, 28, 28)], dim=3)
        save_image(x_concat, os.path.join(sample_dir, 'reconst-{}.png'.format(epoch + 1)))

展示原图像及重构图像

reconsPath = './ave_samples/reconst-30.png'
Image = mpimg.imread(reconsPath)
plt.imshow(Image) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()

显示由潜在空间点Z生成的新图像

genPath = './ave_samples/sampled-30.png'
Image = mpimg.imread(genPath)
plt.imshow(Image) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()