一、SGAN理论基础¶

SGAN将产生式对抗网络（GAN）拓展到半监督学习，通过强制判别器D来输出类别标签。我们在一个数据集上训练一个生成器G 以及一个判别器D，输入是N类当中的一个。在训练的时候，判别器D被用于预测输入是属于 N+1类中的哪一个，这个N+1是对应了生成器G的输出，这里的判别器D同时也充当起了分类器C的效果。这种方法可以用于训练效果更好的判别器D，并且可以比普通的GAN 产生更加高质量的样本。Semi-Supervised GAN有如下优点：

对GANs做了一个新的扩展，允许它同时学习一个生成模型和一个分类器。我们把这个扩展叫做半监督GAN或SGAN
论文实验结果表明，SGAN在有限数据集上比没有生成部分的基准分类器提升了分类性能
论文实验结果表明，SGAN可以显著地提升生成样本的质量并降低生成器的训练时间

通过生成的效果图可以明显发现普通DCGAN算法与SGAN算法性能优劣

二、代码实现¶

1、配置代码¶

In [ ]:

import argparse
import os
import numpy as np
import math

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.nn.functional as F
import torch

# 创建保存生成图像的文件夹
os.makedirs("images", exist_ok=True)

# 使用 argparse 解析命令行参数
parser = argparse.ArgumentParser()
parser.add_argument("--n_epochs", type=int, default=50, help="训练的轮数")
parser.add_argument("--batch_size", type=int, default=64, help="每个批次的样本数量")
parser.add_argument("--lr", type=float, default=0.0002, help="Adam 优化器的学习率")
parser.add_argument("--b1", type=float, default=0.5, help="Adam 优化器的第一个动量衰减参数")
parser.add_argument("--b2", type=float, default=0.999, help="Adam 优化器的第二个动量衰减参数")
parser.add_argument("--n_cpu", type=int, default=8, help="用于批次生成的 CPU 线程数")
parser.add_argument("--latent_dim", type=int, default=100, help="潜在空间的维度")
parser.add_argument("--num_classes", type=int, default=10, help="数据集的类别数")
parser.add_argument("--img_size", type=int, default=32, help="每个图像的尺寸（高度和宽度相等）")
parser.add_argument("--channels", type=int, default=1, help="图像的通道数（灰度图像通道数为 1）")
parser.add_argument("--sample_interval", type=int, default=400, help="图像采样间隔")
opt = parser.parse_args(['--batch_size','8192'])
print(opt)

# 如果 GPU 可用，则使用 CUDA 加速
cuda = True if torch.cuda.is_available() else False

Namespace(n_epochs=50, batch_size=8192, lr=0.0002, b1=0.5, b2=0.999, n_cpu=8, latent_dim=100, num_classes=10, img_size=32, channels=1, sample_interval=400)

2、初始化权重¶

In [ ]:

def weights_init_normal(m):
    classname = m.__class__.__name__
    if classname.find("Conv") != -1:
        torch.nn.init.normal_(m.weight.data, 0.0, 0.02)
    elif classname.find("BatchNorm") != -1:
        torch.nn.init.normal_(m.weight.data, 1.0, 0.02)
        torch.nn.init.constant_(m.bias.data, 0.0)

3、定义算法模型¶

In [ ]:

import torch.nn as nn

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

        # 创建一个标签嵌入层，用于将条件标签映射到潜在空间
        self.label_emb = nn.Embedding(opt.num_classes, opt.latent_dim)

        # 初始化图像尺寸，用于上采样之前
        self.init_size = opt.img_size // 4  # Initial size before upsampling

        # 第一个全连接层，将随机噪声映射到合适的维度
        self.l1 = nn.Sequential(nn.Linear(opt.latent_dim, 128 * self.init_size ** 2))

        # 生成器的卷积块
        self.conv_blocks = nn.Sequential(
            nn.BatchNorm2d(128),
            nn.Upsample(scale_factor=2),
            nn.Conv2d(128, 128, 3, stride=1, padding=1),
            nn.BatchNorm2d(128, 0.8),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Upsample(scale_factor=2),
            nn.Conv2d(128, 64, 3, stride=1, padding=1),
            nn.BatchNorm2d(64, 0.8),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Conv2d(64, opt.channels, 3, stride=1, padding=1),
            nn.Tanh(),
        )

    def forward(self, noise):
        out = self.l1(noise)
        out = out.view(out.shape[0], 128, self.init_size, self.init_size)
        img = self.conv_blocks(out)
        return img

class Discriminator(nn.Module):
    def __init__(self):
        super(Discriminator, self).__init__()

        def discriminator_block(in_filters, out_filters, bn=True):
            """返回每个鉴别器块的层"""
            block = [nn.Conv2d(in_filters, out_filters, 3, 2, 1), nn.LeakyReLU(0.2, inplace=True), nn.Dropout2d(0.25)]
            if bn:
                block.append(nn.BatchNorm2d(out_filters, 0.8))
            return block

        # 鉴别器的卷积块
        self.conv_blocks = nn.Sequential(
            *discriminator_block(opt.channels, 16, bn=False),
            *discriminator_block(16, 32),
            *discriminator_block(32, 64),
            *discriminator_block(64, 128),
        )

        # 下采样图像的高度和宽度
        ds_size = opt.img_size // 2 ** 4

        # 输出层
        self.adv_layer = nn.Sequential(nn.Linear(128 * ds_size ** 2, 1), nn.Sigmoid())  # 用于鉴别真假的输出层
        self.aux_layer = nn.Sequential(nn.Linear(128 * ds_size ** 2, opt.num_classes + 1), nn.Softmax())  # 用于鉴别类别的输出层

    def forward(self, img):
        out = self.conv_blocks(img)
        out = out.view(out.shape[0], -1)
        validity = self.adv_layer(out)
        label = self.aux_layer(out)

        return validity, label

4、配置模型¶

In [ ]:

# 定义损失函数
adversarial_loss = torch.nn.BCELoss()  # 二元交叉熵损失，用于对抗训练
auxiliary_loss = torch.nn.CrossEntropyLoss()  # 交叉熵损失，用于辅助分类

# 初始化生成器和鉴别器
generator = Generator()  # 创建生成器实例
discriminator = Discriminator()  # 创建鉴别器实例

# 如果使用GPU，将模型和损失函数移至GPU上
if cuda:
    generator.cuda()
    discriminator.cuda()
    adversarial_loss.cuda()
    auxiliary_loss.cuda()

# 初始化模型权重
generator.apply(weights_init_normal)  # 初始化生成器的权重
discriminator.apply(weights_init_normal)  # 初始化鉴别器的权重

# 配置数据加载器
os.makedirs("./data/mnist_G7", exist_ok=True)  # 创建存储MNIST数据集的文件夹
dataloader = torch.utils.data.DataLoader(
    datasets.MNIST(
        "./data/mnist_G7",
        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,
)

# 优化器
optimizer_G = torch.optim.Adam(generator.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))  # 生成器的优化器
optimizer_D = torch.optim.Adam(discriminator.parameters(), lr=opt.lr, betas=(opt.b1, opt.b2))  # 鉴别器的优化器

# 根据是否使用GPU选择数据类型
FloatTensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
LongTensor = torch.cuda.LongTensor if cuda else torch.LongTensor

Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz
Using downloaded and verified file: ./data/mnist_G7\MNIST\raw\train-images-idx3-ubyte.gz
Extracting ./data/mnist_G7\MNIST\raw\train-images-idx3-ubyte.gz to ./data/mnist_G7\MNIST\raw

Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz
Using downloaded and verified file: ./data/mnist_G7\MNIST\raw\train-labels-idx1-ubyte.gz
Extracting ./data/mnist_G7\MNIST\raw\train-labels-idx1-ubyte.gz to ./data/mnist_G7\MNIST\raw

Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
Using downloaded and verified file: ./data/mnist_G7\MNIST\raw\t10k-images-idx3-ubyte.gz
Extracting ./data/mnist_G7\MNIST\raw\t10k-images-idx3-ubyte.gz to ./data/mnist_G7\MNIST\raw

Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz
Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to ./data/mnist_G7\MNIST\raw\t10k-labels-idx1-ubyte.gz

100%|██████████████████████████████████████████████████████████████████████████████████████| 4542/4542 [00:00<?, ?it/s]

Extracting ./data/mnist_G7\MNIST\raw\t10k-labels-idx1-ubyte.gz to ./data/mnist_G7\MNIST\raw

5、训练模型¶

In [ ]:

for epoch in range(opt.n_epochs):
    for i, (imgs, labels) in enumerate(dataloader):

        batch_size = imgs.shape[0]

        # 定义对抗训练的标签
        valid = Variable(FloatTensor(batch_size, 1).fill_(1.0), requires_grad=False)  # 用于真实样本
        fake = Variable(FloatTensor(batch_size, 1).fill_(0.0), requires_grad=False)  # 用于生成样本
        fake_aux_gt = Variable(LongTensor(batch_size).fill_(opt.num_classes), requires_grad=False)  # 用于生成样本的类别标签

        # 配置输入数据
        real_imgs = Variable(imgs.type(FloatTensor))  # 真实图像
        labels = Variable(labels.type(LongTensor))  # 真实类别标签

        # -----------------
        #  训练生成器
        # -----------------

        optimizer_G.zero_grad()

        # 采样噪声和类别标签作为生成器的输入
        z = Variable(FloatTensor(np.random.normal(0, 1, (batch_size, opt.latent_dim))))

        # 生成一批图像
        gen_imgs = generator(z)

        # 计算生成器的损失，衡量生成器欺骗鉴别器的能力
        validity, _ = discriminator(gen_imgs)
        g_loss = adversarial_loss(validity, valid)

        g_loss.backward()
        optimizer_G.step()

        # ---------------------
        #  训练鉴别器
        # ---------------------

        optimizer_D.zero_grad()

        # 真实图像的损失
        real_pred, real_aux = discriminator(real_imgs)
        d_real_loss = (adversarial_loss(real_pred, valid) + auxiliary_loss(real_aux, labels)) / 2

        # 生成图像的损失
        fake_pred, fake_aux = discriminator(gen_imgs.detach())
        d_fake_loss = (adversarial_loss(fake_pred, fake) + auxiliary_loss(fake_aux, fake_aux_gt)) / 2

        # 总的鉴别器损失
        d_loss = (d_real_loss + d_fake_loss) / 2

        # 计算鉴别器准确率
        pred = np.concatenate([real_aux.data.cpu().numpy(), fake_aux.data.cpu().numpy()], axis=0)
        gt = np.concatenate([labels.data.cpu().numpy(), fake_aux_gt.data.cpu().numpy()], axis=0)
        d_acc = np.mean(np.argmax(pred, axis=1) == gt)

        d_loss.backward()
        optimizer_D.step()

        batches_done = epoch * len(dataloader) + i
        if batches_done % opt.sample_interval == 0:
            save_image(gen_imgs.data[:25], "images/%d.png" % batches_done, nrow=5, normalize=True)

    print(
        "[Epoch %d/%d] [Batch %d/%d] [D loss: %f, acc: %d%%] [G loss: %f]"
        % (epoch, opt.n_epochs, i, len(dataloader), d_loss.item(), 100 * d_acc, g_loss.item())
    )

D:\Code\Anaconda\envs\dl\lib\site-packages\torch\nn\modules\container.py:217: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.
  input = module(input)

[Epoch 0/50] [Batch 7/8] [D loss: 1.544445, acc: 41%] [G loss: 0.692882]
[Epoch 1/50] [Batch 7/8] [D loss: 1.543716, acc: 49%] [G loss: 0.682594]
[Epoch 2/50] [Batch 7/8] [D loss: 1.540643, acc: 50%] [G loss: 0.680204]
[Epoch 3/50] [Batch 7/8] [D loss: 1.535141, acc: 50%] [G loss: 0.649317]
[Epoch 4/50] [Batch 7/8] [D loss: 1.516119, acc: 49%] [G loss: 0.558261]
[Epoch 5/50] [Batch 7/8] [D loss: 1.432424, acc: 50%] [G loss: 0.559069]
[Epoch 6/50] [Batch 7/8] [D loss: 1.389404, acc: 50%] [G loss: 0.632853]
[Epoch 7/50] [Batch 7/8] [D loss: 1.380128, acc: 50%] [G loss: 0.626831]
[Epoch 8/50] [Batch 7/8] [D loss: 1.369062, acc: 50%] [G loss: 0.677921]
[Epoch 9/50] [Batch 7/8] [D loss: 1.370481, acc: 50%] [G loss: 0.664256]
[Epoch 10/50] [Batch 7/8] [D loss: 1.374016, acc: 50%] [G loss: 0.635747]
[Epoch 11/50] [Batch 7/8] [D loss: 1.371492, acc: 50%] [G loss: 0.678369]
[Epoch 12/50] [Batch 7/8] [D loss: 1.368379, acc: 50%] [G loss: 0.700721]
[Epoch 13/50] [Batch 7/8] [D loss: 1.368315, acc: 50%] [G loss: 0.657885]
[Epoch 14/50] [Batch 7/8] [D loss: 1.376032, acc: 50%] [G loss: 0.690881]
[Epoch 15/50] [Batch 7/8] [D loss: 1.366467, acc: 50%] [G loss: 0.727538]
[Epoch 16/50] [Batch 7/8] [D loss: 1.369054, acc: 50%] [G loss: 0.678619]
[Epoch 17/50] [Batch 7/8] [D loss: 1.362761, acc: 50%] [G loss: 0.656075]
[Epoch 18/50] [Batch 7/8] [D loss: 1.374287, acc: 50%] [G loss: 0.685217]
[Epoch 19/50] [Batch 7/8] [D loss: 1.374793, acc: 50%] [G loss: 0.727006]
[Epoch 20/50] [Batch 7/8] [D loss: 1.366290, acc: 50%] [G loss: 0.730007]
[Epoch 21/50] [Batch 7/8] [D loss: 1.367087, acc: 50%] [G loss: 0.690608]
[Epoch 22/50] [Batch 7/8] [D loss: 1.369826, acc: 50%] [G loss: 0.642754]
[Epoch 23/50] [Batch 7/8] [D loss: 1.368557, acc: 50%] [G loss: 0.681226]
[Epoch 24/50] [Batch 7/8] [D loss: 1.374677, acc: 50%] [G loss: 0.709752]
[Epoch 25/50] [Batch 7/8] [D loss: 1.368476, acc: 50%] [G loss: 0.722963]
[Epoch 26/50] [Batch 7/8] [D loss: 1.368087, acc: 50%] [G loss: 0.695970]
[Epoch 27/50] [Batch 7/8] [D loss: 1.371420, acc: 50%] [G loss: 0.656335]
[Epoch 28/50] [Batch 7/8] [D loss: 1.367887, acc: 50%] [G loss: 0.678238]
[Epoch 29/50] [Batch 7/8] [D loss: 1.367137, acc: 50%] [G loss: 0.704793]
[Epoch 30/50] [Batch 7/8] [D loss: 1.372444, acc: 50%] [G loss: 0.699853]
[Epoch 31/50] [Batch 7/8] [D loss: 1.370048, acc: 50%] [G loss: 0.692634]
[Epoch 32/50] [Batch 7/8] [D loss: 1.365426, acc: 50%] [G loss: 0.697769]
[Epoch 33/50] [Batch 7/8] [D loss: 1.370704, acc: 50%] [G loss: 0.676067]
[Epoch 34/50] [Batch 7/8] [D loss: 1.370108, acc: 50%] [G loss: 0.673559]
[Epoch 35/50] [Batch 7/8] [D loss: 1.367148, acc: 50%] [G loss: 0.693939]
[Epoch 36/50] [Batch 7/8] [D loss: 1.372817, acc: 50%] [G loss: 0.691008]
[Epoch 37/50] [Batch 7/8] [D loss: 1.370204, acc: 50%] [G loss: 0.693787]
[Epoch 38/50] [Batch 7/8] [D loss: 1.365855, acc: 50%] [G loss: 0.705938]
[Epoch 39/50] [Batch 7/8] [D loss: 1.367928, acc: 50%] [G loss: 0.707550]
[Epoch 40/50] [Batch 7/8] [D loss: 1.369588, acc: 50%] [G loss: 0.693974]
[Epoch 41/50] [Batch 7/8] [D loss: 1.371375, acc: 50%] [G loss: 0.666972]
[Epoch 42/50] [Batch 7/8] [D loss: 1.366990, acc: 50%] [G loss: 0.679381]
[Epoch 43/50] [Batch 7/8] [D loss: 1.370401, acc: 50%] [G loss: 0.696619]
[Epoch 44/50] [Batch 7/8] [D loss: 1.369853, acc: 50%] [G loss: 0.704175]
[Epoch 45/50] [Batch 7/8] [D loss: 1.366559, acc: 50%] [G loss: 0.708872]
[Epoch 46/50] [Batch 7/8] [D loss: 1.366858, acc: 50%] [G loss: 0.694374]
[Epoch 47/50] [Batch 7/8] [D loss: 1.370696, acc: 50%] [G loss: 0.662108]
[Epoch 48/50] [Batch 7/8] [D loss: 1.371429, acc: 50%] [G loss: 0.682755]
[Epoch 49/50] [Batch 7/8] [D loss: 1.368934, acc: 50%] [G loss: 0.710603]