pytorch GAN伪造手写体mnist数据集方式

一，mnist数据集

形如上图的数字手写体就是mnist数据集。

二，GAN原理(生成对抗网络)

GAN网络一共由两部分组成：一个是伪造器(Generator，简称G)，一个是判别器(Discrimniator，简称D)

一开始，G由服从某几个分布(如高斯分布)的噪音组成，生成的图片不断送给D判断是否正确，直到G生成的图片连D都判断以为是真的。D每一轮除了看过G生成的假图片以外，还要见数据集中的真图片，以前者和后者得到的损失函数值为依据更新D网络中的权值。因此G和D都在不停地更新权值。以下图为例：

在v1时的G只不过是 一堆噪声，见过数据集(real images)的D肯定能判断出G所生成的是假的。当然G也能知道D判断它是假的这个结果，因此G就会更新权值，到v2的时候，G就能生成更逼真的图片来让D判断，当然在v2时D也是会先看一次真图片，再去判断G所生成的图片。以此类推，不断循环就是GAN的思想。

三，训练代码

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

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=8, 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("--img_size", type=int, default=28, 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="interval betwen image samples")
opt = parser.parse_args()
print(opt)

img_shape = (opt.channels, opt.img_size, opt.img_size) # 确定图片输入的格式为(1，28，28)，由于mnist数据集是灰度图所以通道为1
cuda = True if torch.cuda.is_available() else False

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

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

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

# Initialize generator and discriminator
generator = Generator()
discriminator = Discriminator()

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

# Configure data loader
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,
)

# 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 cuda else torch.FloatTensor

# ----------
# Training
# ----------
if __name__ == '__main__':
for epoch in range(opt.n_epochs):
 for i, (imgs, _) in enumerate(dataloader):
  # print(imgs.shape)
  # Adversarial ground truths
  valid = Variable(Tensor(imgs.size(0), 1).fill_(1.0), requires_grad=False) # 全1
  fake = Variable(Tensor(imgs.size(0), 1).fill_(0.0), requires_grad=False) # 全0
  # Configure input
  real_imgs = Variable(imgs.type(Tensor))

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

optimizer_G.zero_grad() # 清空G网络 上一个batch的梯度

# Sample noise as generator input
  z = Variable(Tensor(np.random.normal(0, 1, (imgs.shape[0], opt.latent_dim)))) # 生成的噪音，均值为0方差为1维度为(64，100)的噪音
  # Generate a batch of images
  gen_imgs = generator(z)
  # Loss measures generator's ability to fool the discriminator
  g_loss = adversarial_loss(discriminator(gen_imgs), valid)

g_loss.backward() # g_loss用于更新G网络的权值，g_loss于D网络的判断结果 有关
  optimizer_G.step()

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

optimizer_D.zero_grad() # 清空D网络 上一个batch的梯度
  # Measure discriminator's ability to classify real from generated samples
  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() # d_loss用于更新D网络的权值
  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:
   save_image(gen_imgs.data[:25], "images/％d.png" ％ batches_done, nrow=5, normalize=True) # 保存一个batchsize中的25张
  if (epoch+1) ％2 ==0:
   print('save..')
   torch.save(generator,'g％d.pth' ％ epoch)
   torch.save(discriminator,'d％d.pth' ％ epoch)

运行结果：

一开始时，G生成的全是杂音：

然后逐渐呈现数字的雏形：

最后一次生成的结果：

四，测试代码：

导入最后保存生成器的模型：

from gan import Generator,Discriminator
import torch
import matplotlib.pyplot as plt
from torch.autograd import Variable
import numpy as np
from torchvision.utils import save_image

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
Tensor = torch.cuda.FloatTensor
g = torch.load('g199.pth') #导入生成器Generator模型
#d = torch.load('d.pth')
g = g.to(device)
#d = d.to(device)

z = Variable(Tensor(np.random.normal(0, 1, (64, 100)))) #输入的噪音
gen_imgs =g(z) #生产图片
save_image(gen_imgs.data[:25], "images.png" , nrow=5, normalize=True)

生成结果：

来源：https://blog.csdn.net/u014453898/article/details/95044228