服务器之家

服务器之家 > 正文

pytorch:实现简单的GAN示例(MNIST数据集)

时间:2020-04-29 10:18     来源/作者:xckkcxxck

我就废话不多说了,直接上代码吧!

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
# -*- coding: utf-8 -*-
"""
Created on Sat Oct 13 10:22:45 2018
@author: www
"""
 
import torch
from torch import nn
from torch.autograd import Variable
 
import torchvision.transforms as tfs
from torch.utils.data import DataLoader, sampler
from torchvision.datasets import MNIST
 
import numpy as np
 
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
 
plt.rcParams['figure.figsize'] = (10.0, 8.0) # 设置画图的尺寸
plt.rcParams['image.interpolation'] = 'nearest'
plt.rcParams['image.cmap'] = 'gray'
 
def show_images(images): # 定义画图工具
  images = np.reshape(images, [images.shape[0], -1])
  sqrtn = int(np.ceil(np.sqrt(images.shape[0])))
  sqrtimg = int(np.ceil(np.sqrt(images.shape[1])))
 
  fig = plt.figure(figsize=(sqrtn, sqrtn))
  gs = gridspec.GridSpec(sqrtn, sqrtn)
  gs.update(wspace=0.05, hspace=0.05)
 
  for i, img in enumerate(images):
    ax = plt.subplot(gs[i])
    plt.axis('off')
    ax.set_xticklabels([])
    ax.set_yticklabels([])
    ax.set_aspect('equal')
    plt.imshow(img.reshape([sqrtimg,sqrtimg]))
  return
  
def preprocess_img(x):
  x = tfs.ToTensor()(x)
  return (x - 0.5) / 0.5
 
def deprocess_img(x):
  return (x + 1.0) / 2.0
 
class ChunkSampler(sampler.Sampler): # 定义一个取样的函数
  """Samples elements sequentially from some offset.
  Arguments:
    num_samples: # of desired datapoints
    start: offset where we should start selecting from
  """
  def __init__(self, num_samples, start=0):
    self.num_samples = num_samples
    self.start = start
 
  def __iter__(self):
    return iter(range(self.start, self.start + self.num_samples))
 
  def __len__(self):
    return self.num_samples
    
NUM_TRAIN = 50000
NUM_VAL = 5000
 
NOISE_DIM = 96
batch_size = 128
 
train_set = MNIST('E:/data', train=True, transform=preprocess_img)
 
train_data = DataLoader(train_set, batch_size=batch_size, sampler=ChunkSampler(NUM_TRAIN, 0))
 
val_set = MNIST('E:/data', train=True, transform=preprocess_img)
 
val_data = DataLoader(val_set, batch_size=batch_size, sampler=ChunkSampler(NUM_VAL, NUM_TRAIN))
 
imgs = deprocess_img(train_data.__iter__().next()[0].view(batch_size, 784)).numpy().squeeze() # 可视化图片效果
show_images(imgs)
 
#判别网络
def discriminator():
  net = nn.Sequential(   
      nn.Linear(784, 256),
      nn.LeakyReLU(0.2),
      nn.Linear(256, 256),
      nn.LeakyReLU(0.2),
      nn.Linear(256, 1)
    )
  return net
  
#生成网络
def generator(noise_dim=NOISE_DIM): 
  net = nn.Sequential(
    nn.Linear(noise_dim, 1024),
    nn.ReLU(True),
    nn.Linear(1024, 1024),
    nn.ReLU(True),
    nn.Linear(1024, 784),
    nn.Tanh()
  )
  return net
  
#判别器的 loss 就是将真实数据的得分判断为 1,假的数据的得分判断为 0,而生成器的 loss 就是将假的数据判断为 1
 
bce_loss = nn.BCEWithLogitsLoss()#交叉熵损失函数
 
def discriminator_loss(logits_real, logits_fake): # 判别器的 loss
  size = logits_real.shape[0]
  true_labels = Variable(torch.ones(size, 1)).float()
  false_labels = Variable(torch.zeros(size, 1)).float()
  loss = bce_loss(logits_real, true_labels) + bce_loss(logits_fake, false_labels)
  return loss
  
def generator_loss(logits_fake): # 生成器的 loss
  size = logits_fake.shape[0]
  true_labels = Variable(torch.ones(size, 1)).float()
  loss = bce_loss(logits_fake, true_labels)
  return loss
  
# 使用 adam 来进行训练,学习率是 3e-4, beta1 是 0.5, beta2 是 0.999
def get_optimizer(net):
  optimizer = torch.optim.Adam(net.parameters(), lr=3e-4, betas=(0.5, 0.999))
  return optimizer
  
def train_a_gan(D_net, G_net, D_optimizer, G_optimizer, discriminator_loss, generator_loss, show_every=250,
        noise_size=96, num_epochs=10):
  iter_count = 0
  for epoch in range(num_epochs):
    for x, _ in train_data:
      bs = x.shape[0]
      # 判别网络
      real_data = Variable(x).view(bs, -1) # 真实数据
      logits_real = D_net(real_data) # 判别网络得分
      
      sample_noise = (torch.rand(bs, noise_size) - 0.5) / 0.5 # -1 ~ 1 的均匀分布
      g_fake_seed = Variable(sample_noise)
      fake_images = G_net(g_fake_seed) # 生成的假的数据
      logits_fake = D_net(fake_images) # 判别网络得分
 
      d_total_error = discriminator_loss(logits_real, logits_fake) # 判别器的 loss
      D_optimizer.zero_grad()
      d_total_error.backward()
      D_optimizer.step() # 优化判别网络
      
      # 生成网络
      g_fake_seed = Variable(sample_noise)
      fake_images = G_net(g_fake_seed) # 生成的假的数据
 
      gen_logits_fake = D_net(fake_images)
      g_error = generator_loss(gen_logits_fake) # 生成网络的 loss
      G_optimizer.zero_grad()
      g_error.backward()
      G_optimizer.step() # 优化生成网络
 
      if (iter_count % show_every == 0):
        print('Iter: {}, D: {:.4}, G:{:.4}'.format(iter_count, d_total_error.item(), g_error.item()))
        imgs_numpy = deprocess_img(fake_images.data.cpu().numpy())
        show_images(imgs_numpy[0:16])
        plt.show()
        print()
      iter_count += 1
 
D = discriminator()
G = generator()
 
D_optim = get_optimizer(D)
G_optim = get_optimizer(G)
 
train_a_gan(D, G, D_optim, G_optim, discriminator_loss, generator_loss)     

以上这篇pytorch:实现简单的GAN示例(MNIST数据集)就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/xckkcxxck/article/details/83037025

相关文章

热门资讯

2022年最旺的微信头像大全 微信头像2022年最新版图片
2022年最旺的微信头像大全 微信头像2022年最新版图片 2022-01-10
蜘蛛侠3英雄无归3正片免费播放 蜘蛛侠3在线观看免费高清完整
蜘蛛侠3英雄无归3正片免费播放 蜘蛛侠3在线观看免费高清完整 2021-08-24
背刺什么意思 网络词语背刺是什么梗
背刺什么意思 网络词语背刺是什么梗 2020-05-22
yue是什么意思 网络流行语yue了是什么梗
yue是什么意思 网络流行语yue了是什么梗 2020-10-11
暖暖日本高清免费中文 暖暖在线观看免费完整版韩国
暖暖日本高清免费中文 暖暖在线观看免费完整版韩国 2021-05-08
返回顶部