【GAN实战】MNIST手写数字生成——基于pytorch的GAN-MNIST

MNIST数据集介绍

首先介绍一下MNIST数据集，MNIST是个手写数字图片集，每张图片都做了归一化处理，大小是28x28，并且是灰度图像，所以shape是1x28x28
官方网址：http://yann.lecun.com/exdb/mnist/，来到数据集下载页面，可以看到：

主要包括四个文件：

类别	文件名	描述
训练集图片	train-images-idx3-ubyte.gz（9.9M）	包含60000个样本
训练集标签	train-labels-idx1-ubyte.gz（29KB）	包含60000个标签
测试集图片	t10k-images-idx3-ubyte.gz（1.6M）	包含10000个样本
测试集标签	t10k-labels-idx1-ubyte.gz（5KB）	包含10000个样本

解压后，得到的并不是标准的图像格式，想深入了解的可以去看官网的描述。要把数据读取出来也得按照文件数据结构来，可以参考这篇博客的代码。

在torchvison.datasets中已经为我们封装好了MNIST数据集（解压+读取）

在论文Gradient-Based Learning Applied to Document Recognition，作者首次提出了LeNet-5 网络，利用上述数据集实现了手写字体的识别，网络模型如下：

GAN-MNIST

用GAN生成MNIST图片

0.准备工作：导入相关包&参数设置

0.1 导入相关包

import os
import numpy as np
import math

# 命令行参数解析包
import argparse

# 导入pytorch相关包
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

0.2 命令行参数配置

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("--latent_dim", type=int, default=100, help="dimensionality of the latent space")
parser.add_argument("--sample_interval", type=int, default=400, help="interval between image samples")
opt = parser.parse_args()
print(opt)

img_shape = (1, 28, 28) # 图像的尺寸
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)

输出：

# output：
Namespace(b1=0.5, b2=0.999, batch_size=64, latent_dim=100, lr=0.0002, n_epochs=200, sample_interval=400)
cuda

1.数据预处理（transforms等）

tfm = transforms.Compose([transforms.Resize(28),
                          transforms.ToTensor(),
                          transforms.Normalize([0.5],[0.5])])

2.数据读取（Dataset和Dataloader）

data_loader = torch.utils.data.DataLoader(
    datasets.MNIST("./data", train=True, download=True, transform=tfm),
    batch_size=opt.batch_size,
    shuffle=True)

3.生成器和判别器的网络搭建

class Generator(nn.Module):
    def __init__(self):
        super(Generator, self).__init__()
         # 定义一个block，每个block的操作就是线性层，正则化（可选），再激活层
        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(opt.latent_dim, 128, normalize=False),
            *block(128, 256),
            *block(256, 512),
            *block(512, 1024),
            nn.Linear(1024, int(np.prod(img_shape))),
            nn.Tanh()
        )

    def forward(self, z):
        img = self.model(z)
        img = img.view(img.size(0), *img_shape)
        return img


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

        self.model = nn.Sequential(
            nn.Linear(int(np.prod(img_shape)), 512),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Linear(512, 256),
            nn.LeakyReLU(0.2, inplace=True),
            nn.Linear(256, 1),
            nn.Sigmoid(),
        )

    def forward(self, img):
        img_flat = img.view(img.size(0), -1)
        validity = self.model(img_flat)

        return validity

4.开始训练

4.1 loss函数和优化器

if __name__ == '__main__':
    # Initialize generator and discriminator
    generator = Generator().to(device)
    discriminator = Discriminator().to(device)

    # Loss function
    adversarial_loss = torch.nn.BCELoss().to(device)

    # Optimizers
    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))

    Tensor = torch.cuda.FloatTensor if torch.cuda.is_available() else torch.FloatTensor
    # 创建存储生成图片的目录
    os.makedirs("images", exist_ok=True)

4.2 训练集上开始训练

    for epoch in range(opt.n_epochs):
        for i, (imgs, _) in enumerate(data_loader):

            # Adversarial ground truths，valid:64x1的全1向量，fake:64x1的全零向量
            valid = Variable(Tensor(imgs.size(0), 1).fill_(1.0), requires_grad=False)
            fake = Variable(Tensor(imgs.size(0), 1).fill_(0.0), requires_grad=False)

            # Configure input ,让输入图像适应网络
            real_imgs = Variable(imgs.type(Tensor))

            # -----------------
            #  Train Generator
            # -----------------

            optimizer_G.zero_grad()

            # Sample noise as generator input,z是有标准正态分布采样得到的64x100向量
            z = Variable(Tensor(np.random.normal(0, 1, (imgs.shape[0], opt.latent_dim))))

            # Generate a batch of images,64x1x28x28
            gen_imgs = generator(z)

            # Loss measures generator's ability to fool the discriminator
            """生成器损失函数：让生成的图像通过判别器后与valid相比较，越靠近越好"""
            g_loss = adversarial_loss(discriminator(gen_imgs), valid)

            g_loss.backward()
            optimizer_G.step()

            # ---------------------
            #  Train Discriminator
            # ---------------------

            optimizer_D.zero_grad()

            # Measure discriminator's ability to classify real from generated samples
            """
            判别器损失函数：一方面让真实图片通过判别器与valid越接近，
            另一方面让生成的图片通过判别器与fake越接近(正好与生成器相矛盾，这样才能提高判别能力)
            """
            real_loss = adversarial_loss(discriminator(real_imgs), valid)
            fake_loss = adversarial_loss(discriminator(gen_imgs.detach()), fake)
            d_loss = (real_loss + 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(data_loader), d_loss.item(), g_loss.item())
            )

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

输出：

# output：
[Epoch 0/200] [Batch 0/938] [D loss: 0.712640] [G loss: 0.712526]
[Epoch 0/200] [Batch 1/938] [D loss: 0.622424] [G loss: 0.709658]
[Epoch 0/200] [Batch 2/938] [D loss: 0.554259] [G loss: 0.707226]
[Epoch 0/200] [Batch 3/938] [D loss: 0.495899] [G loss: 0.704608]
[Epoch 0/200] [Batch 4/938] [D loss: 0.452283] [G loss: 0.701480]
[Epoch 0/200] [Batch 5/938] [D loss: 0.419015] [G loss: 0.696898]
[Epoch 0/200] [Batch 6/938] [D loss: 0.396711] [G loss: 0.692087]
[Epoch 0/200] [Batch 7/938] [D loss: 0.384795] [G loss: 0.685381]
[Epoch 0/200] [Batch 8/938] [D loss: 0.378744] [G loss: 0.675533]
[Epoch 0/200] [Batch 9/938] [D loss: 0.378815] [G loss: 0.665169]
[Epoch 0/200] [Batch 10/938] [D loss: 0.382517] [G loss: 0.653243]
…………

在images/目录下，会看到生成的图片逐渐趋向真实的MNIST数据集……