Pytorch CGAN实现MNIST手写数字数据集

简介

生成对抗网络（Generative Adversarial Networks，简称GAN）是一种深度学习模型，通过生成器和判别器的对抗训练，从随机噪声中生成逼真的数据。在本博客中，我们将使用PyTorch框架实现一个条件生成对抗网络（Conditional GAN，简称CGAN），并利用MNIST数据集进行手写数字的生成。

项目概述

在这个项目中，我们将实现一个生成器（Generator）和一个判别器（Discriminator）。生成器将随机噪声和条件标签作为输入，生成逼真的手写数字图像。判别器则负责判断输入图像是真实的MNIST图像还是生成器生成的假图像。

条件生成对抗网络（CGAN）原理

CGAN是对标准GAN的扩展，引入了条件信息。在生成器和判别器的训练中，额外的条件信息被引入，使得生成的数据能够受到指定条件的控制。在MNIST数据集的例子中，这个条件信息可以是手写数字的标签。这意味着我们可以生成特定数字的图像。

生成器（Generator）

生成器的目标是将随机噪声和条件标签转化为逼真的图像。在CGAN中，生成器通常包括一个嵌入层（embedding layer）用于处理条件标签，以及一系列全连接层和批量归一化层，最终输出生成的图像。生成器的架构可以根据具体任务和数据集进行调整。

判别器（Discriminator）

判别器的任务是判断输入的图像是真实的还是生成器生成的假图像。与生成器一样，判别器也包含一个嵌入层用于处理条件标签，并通过全连接层进行判别。通过引入条件信息，判别器可以更好地识别生成的图像的真实性。

损失函数和优化器

在CGAN中，使用二元交叉熵损失函数（Binary Cross Entropy Loss）来衡量生成器生成图像的逼真程度以及判别器的判别能力。优化器使用Adam优化算法

完整训练代码如下：

import argparse
import os
import numpy as np

import torchvision.transforms as transforms
from torchvision.utils import save_image

from torch.utils.data import DataLoader
from torchvision import datasets
from torch.autograd import Variable

import torch.nn as nn
import torch

os.makedirs("images", exist_ok=True)

parser = argparse.ArgumentParser()
parser.add_argument("--n_epochs", type=int, default=200, help="number of epochs of training")
parser.add_argument("--batch_size", type=int, default=64, help="size of the batches")
parser.add_argument("--lr", type=float, default=0.0002, help="adam: learning rate")
parser.add_argument("--b1", type=float, default=0.5, help="adam: decay of first order momentum of gradient")
parser.add_argument("--b2", type=float, default=0.999, help="adam: decay of first order momentum of gradient")
parser.add_argument("--n_cpu", type=int, default=4, help="number of cpu threads to use during batch generation")
parser.add_argument("--latent_dim", type=int, default=100, help="dimensionality of the latent space")
parser.add_argument("--n_classes", type=int, default=10, help="number of classes for dataset")
parser.add_argument("--img_size", type=int, default=32, help="size of each image dimension")
parser.add_argument("--channels", type=int, default=1, help="number of image channels")
parser.add_argument("--sample_interval", type=int, default=400, help="图像采样间隔")
parser.add_argument("--input_shape", type=tuple, default=(1, 32, 32), help="输入图像的尺寸")
parser.add_argument("--input_dim", type=int, default=100, help="inpput dimension")
parser.add_argument("--class_nummber", type=int, default=10, help="classes number")
opt = parser.parse_args()
print(opt)

img_shape = (opt.channels, opt.img_size, opt.img_size)

cuda = True if torch.cuda.is_available() else False


class Generator(nn.Module):
    def __init__(self, input_dim, n_classes, img_shape):
        super(Generator, self).__init__()

        self.input_dim = input_dim
        self.n_classes = n_classes
        self.img_shape = img_shape
        self.label_emb = nn.Embedding(self.n_classes, self.n_classes)

        def block(in_feat, out_feat, normalize=True):
            layers = [nn.Linear(in_feat, out_feat)]
            if normalize:
                layers.append(nn.BatchNorm1d(out_feat, 0.8))
            layers.append(nn.LeakyReLU(0.2, inplace=True))
            return layers

        self.model = nn.Sequential(
            *block(self.input_dim + self.n_classes, 128, normalize=False),
            *block(128, 256),
            *block(256, 512),
            *block(512, 1024),
            nn.Linear(1024, int(np.prod(self.img_shape))),
            nn.Tanh()
        )

    def forward(self, noise, labels):
        gen_input = torch.cat((self.label_emb(labels), noise), -1)
        img = self.model(gen_input)
        img = img.view(img.size(0), *self.img_shape)
        return img


class Discriminator(nn.Module):
    def __init__(self, n_classes, img_shape):
        super(Discriminator, self).__init__()
        self.n_classes = n_classes
        self.img_shape = img_shape
        self.label_embedding = nn.Embedding(self.n_classes, self.n_classes)

        self.model = nn.Sequential(
            nn.Linear(self.n_classes + int(np.prod(self.img_shape)), 512),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Linear(512, 512),
            nn.Dropout(0.4),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Linear(512, 512),
            nn.Dropout(0.4),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Linear(512, 1),
            nn.Sigmoid()
        )

    def forward(self, img, labels):
        d_in = torch.cat((img.view(img.size(0), -1), self.label_embedding(labels)), -1)
        validity = self.model(d_in)
        return validity


os.makedirs("data/mnist", exist_ok=True)
dataloader = torch.utils.data.DataLoader(
    datasets.MNIST(
        "data/mnist",
        train=True,
        download=True,
        transform=transforms.Compose(
            [transforms.Resize(opt.img_size), transforms.ToTensor(), transforms.Normalize([0.5], [0.5])]
        ),
    ),
    batch_size=opt.batch_size,
    shuffle=True,
)

adversarial_loss = torch.nn.BCELoss()
generator = Generator(input_dim=opt.input_dim, n_classes=opt.n_classes, img_shape=opt.input_shape)
discriminator = Discriminator(n_classes=opt.n_classes, img_shape=opt.input_shape)

if cuda:
    generator.cuda()
    discriminator.cuda()
    adversarial_loss.cuda()

optimizer_G = torch.optim.Adam(generator.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))

os.makedirs("cgan-images", exist_ok=True)
os.makedirs("saved_models", exist_ok=True)

FloatTensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
LongTensor = torch.cuda.LongTensor if cuda else torch.LongTensor

for epoch in range(opt.n_epochs):
    for i, (imgs, labels) in enumerate(dataloader):
        batch_size = imgs.shape[0]

        valid = Variable(FloatTensor(batch_size, 1).fill_(1.0), requires_grad=False)
        fake = Variable(FloatTensor(batch_size, 1).fill_(0.0), requires_grad=False)

        real_imgs = Variable(imgs.type(FloatTensor))
        labels = Variable(labels.type(LongTensor))

        optimizer_G.zero_grad()

        z = Variable(FloatTensor(np.random.normal(0, 1, (batch_size, opt.latent_dim))))  # 噪声样本
        gen_labels = Variable(LongTensor(np.random.randint(0, opt.n_classes, batch_size)))  # 对应的标签

        gen_imgs = generator(z, gen_labels)

        validity = discriminator(gen_imgs, gen_labels)
        g_loss = adversarial_loss(validity, valid)
        g_loss.backward()
        optimizer_G.step()

        optimizer_D.zero_grad()

        validity_real = discriminator(real_imgs, labels)
        d_real_loss = adversarial_loss(validity_real, valid)
        validity_fake = discriminator(gen_imgs.detach(), gen_labels)
        d_fake_loss = adversarial_loss(validity_fake, fake)

        d_loss = (d_real_loss + d_fake_loss) / 2
        d_loss.backward()
        optimizer_D.step()

        print(
            "[Epoch %d/%d] [Batch %d/%d] [D loss: %f] [G loss: %f]"
            % (epoch, opt.n_epochs, i, len(dataloader), d_loss.item(), g_loss.item())
        )

        batches_done = epoch * len(dataloader) + i
        if batches_done % opt.sample_interval == 0:
            z = Variable(FloatTensor(np.random.normal(0, 1, (10 ** 2, opt.latent_dim))))

            labels = np.array([num for _ in range(10) for num in range(10)])
            labels = Variable(LongTensor(labels))
            gen_imgs = generator(z, labels)
            save_image(gen_imgs.data, "images/%d.png" % batches_done, nrow=10, normalize=True)

    torch.save(generator.state_dict(), "saved_models/generator_best.pth")

测试代码

import torch
import torch.nn as nn
from torchvision.utils import save_image
from torch.autograd import Variable
import numpy as np


class Generator(nn.Module):
    def __init__(self, input_dim, n_classes, img_shape):
        super(Generator, self).__init__()

        self.input_dim = input_dim
        self.n_classes = n_classes
        self.img_shape = img_shape
        self.label_emb = nn.Embedding(self.n_classes, self.n_classes)

        def block(in_feat, out_feat, normalize=True):
            layers = [nn.Linear(in_feat, out_feat)]
            if normalize:
                layers.append(nn.BatchNorm1d(out_feat, 0.8))
            layers.append(nn.LeakyReLU(0.2, inplace=True))
            return layers

        self.model = nn.Sequential(
            *block(self.input_dim + self.n_classes, 128, normalize=False),
            *block(128, 256),
            *block(256, 512),
            *block(512, 1024),
            nn.Linear(1024, int(np.prod(self.img_shape))),
            nn.Tanh()
        )

    def forward(self, noise, labels):
        gen_input = torch.cat((self.label_emb(labels), noise), -1)
        img = self.model(gen_input)
        img = img.view(img.size(0), *self.img_shape)
        return img


generator = Generator(input_dim=100, n_classes=10, img_shape=(1,32,32))
generator.load_state_dict(torch.load("saved_models/generator_best.pth"))

cuda = True if torch.cuda.is_available() else False

if cuda:
    generator.cuda()
generator.eval()

num_images = 100
z = Variable(torch.FloatTensor(np.random.normal(0, 1, (num_images, 100))))
labels = np.array([num for _ in range(10) for num in range(10)])
labels = Variable(torch.LongTensor(labels))

if cuda:
    z = z.cuda()
    labels = labels.cuda()

with torch.no_grad():
    gen_imgs = generator(z, labels)

save_image(gen_imgs.data, "cgan-images/test_generated.png", nrow=10, normalize=True)

print("Generated images saved successfully!")