-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathmain.py
270 lines (189 loc) · 8.72 KB
/
main.py
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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
import torch
import torch.nn as nn
from torchvision.utils import make_grid
import torch.optim as optim
import numpy as np
import torchvision
import torch.nn.functional as F
import matplotlib.pyplot as plt
import torchvision.transforms as transforms
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(device)
tf = transforms.Compose([transforms.Resize(64),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
trainset = torchvision.datasets.CIFAR10(root = './data', train = True, download = True,
transform = tf)
testset = torchvision.datasets.CIFAR10(root = './data', train = False, download = True,
transform = tf)
dataset = torch.utils.data.ConcatDataset([trainset, testset])
trainloader = torch.utils.data.DataLoader(dataset, batch_size = 100,
num_workers = 2, shuffle = True)
print(len(dataset))
print(dataset[0][0].size())
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck', 'fake')
def showImage(images,epoch=-99, idx = -99):
images = images.cpu().numpy()
images = images/2 + 0.5
plt.imshow(np.transpose(images,axes = (1,2,0)))
plt.axis('off')
if epoch!=-99:
plt.savefig("e" + str(epoch) + "i" + str(idx) + ".png")
dataiter = iter(trainloader)
images,labels = dataiter.next()
print(images.size())
showImage(make_grid(images[0:64]))
class Generator(nn.Module):
def __init__(self):
super(Generator,self).__init__()
#input 100*1*1
self.layer1 = nn.Sequential(nn.ConvTranspose2d(100,512,4,1,0,bias = False),
nn.ReLU(True))
#input 512*4*4
self.layer2 = nn.Sequential(nn.ConvTranspose2d(512,256,4,2,1,bias = False),
nn.BatchNorm2d(256),
nn.ReLU(True))
#input 256*8*8
self.layer3 = nn.Sequential(nn.ConvTranspose2d(256,128,4,2,1,bias = False),
nn.BatchNorm2d(128),
nn.ReLU(True))
#input 128*16*16
self.layer4 = nn.Sequential(nn.ConvTranspose2d(128,64,4,2,1,bias = False),
nn.BatchNorm2d(64),
nn.ReLU(True))
#input 64*32*32
self.layer5 = nn.Sequential(nn.ConvTranspose2d(64,3,4,2,1,bias = False),
nn.Tanh())
#output 3*64*64
self.embedding = nn.Embedding(10,100)
def forward(self,noise,label):
label_embedding = self.embedding(label)
x = torch.mul(noise,label_embedding)
x = x.view(-1,100,1,1)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.layer5(x)
return x
class Discriminator(nn.Module):
def __init__(self):
super(Discriminator,self).__init__()
#input 3*64*64
self.layer1 = nn.Sequential(nn.Conv2d(3,64,4,2,1,bias = False),
nn.BatchNorm2d(64),
nn.LeakyReLU(0.2,True),
nn.Dropout2d(0.5))
#input 64*32*32
self.layer2 = nn.Sequential(nn.Conv2d(64,128,4,2,1,bias = False),
nn.BatchNorm2d(128),
nn.LeakyReLU(0.2,True),
nn.Dropout2d(0.5))
#input 128*16*16
self.layer3 = nn.Sequential(nn.Conv2d(128,256,4,2,1,bias = False),
nn.BatchNorm2d(256),
nn.LeakyReLU(0.2,True),
nn.Dropout2d(0.5))
#input 256*8*8
self.layer4 = nn.Sequential(nn.Conv2d(256,512,4,2,1,bias = False),
nn.BatchNorm2d(512),
nn.LeakyReLU(0.2,True))
#input 512*4*4
self.validity_layer = nn.Sequential(nn.Conv2d(512,1,4,1,0,bias = False),
nn.Sigmoid())
self.label_layer = nn.Sequential(nn.Conv2d(512,11,4,1,0,bias = False),
nn.LogSoftmax(dim = 1))
def forward(self,x):
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
validity = self.validity_layer(x)
plabel = self.label_layer(x)
validity = validity.view(-1)
plabel = plabel.view(-1,11)
return validity,plabel
# custom weights initialization called on netG and netD
def weights_init(m):
classname = m.__class__.__name__
if classname.find('Conv') != -1:
m.weight.data.normal_(0.0, 0.02)
elif classname.find('BatchNorm') != -1:
m.weight.data.normal_(1.0, 0.02)
m.bias.data.fill_(0)
gen = Generator().to(device)
gen.apply(weights_init)
disc = Discriminator().to(device)
disc.apply(weights_init)
paramsG = list(gen.parameters())
print(len(paramsG))
paramsD = list(disc.parameters())
print(len(paramsD))
optimG = optim.Adam(gen.parameters(), 0.0002, betas = (0.5,0.999))
optimD = optim.Adam(disc.parameters(), 0.0002, betas = (0.5,0.999))
validity_loss = nn.BCELoss()
real_labels = 0.7 + 0.5 * torch.rand(10, device = device)
fake_labels = 0.3 * torch.rand(10, device = device)
epochs = 10
for epoch in range(1,epochs+1):
for idx, (images,labels) in enumerate(trainloader,0):
batch_size = images.size(0)
labels= labels.to(device)
images = images.to(device)
real_label = real_labels[idx % 10]
fake_label = fake_labels[idx % 10]
fake_class_labels = 10*torch.ones((batch_size,),dtype = torch.long,device = device)
if idx % 25 == 0:
real_label, fake_label = fake_label, real_label
# ---------------------
# disc
# ---------------------
optimD.zero_grad()
# real
validity_label = torch.full((batch_size,),real_label , device = device)
pvalidity, plabels = disc(images)
errD_real_val = validity_loss(pvalidity, validity_label)
errD_real_label = F.nll_loss(plabels,labels)
errD_real = errD_real_val + errD_real_label
errD_real.backward()
D_x = pvalidity.mean().item()
#fake
noise = torch.randn(batch_size,100,device = device)
sample_labels = torch.randint(0,10,(batch_size,),device = device, dtype = torch.long)
fakes = gen(noise,sample_labels)
validity_label.fill_(fake_label)
pvalidity, plabels = disc(fakes.detach())
errD_fake_val = validity_loss(pvalidity, validity_label)
errD_fake_label = F.nll_loss(plabels, fake_class_labels)
errD_fake = errD_fake_val + errD_fake_label
errD_fake.backward()
D_G_z1 = pvalidity.mean().item()
#finally update the params!
errD = errD_real + errD_fake
optimD.step()
# ------------------------
# gen
# ------------------------
optimG.zero_grad()
noise = torch.randn(batch_size,100,device = device)
sample_labels = torch.randint(0,10,(batch_size,),device = device, dtype = torch.long)
validity_label.fill_(1)
fakes = gen(noise,sample_labels)
pvalidity,plabels = disc(fakes)
errG_val = validity_loss(pvalidity, validity_label)
errG_label = F.nll_loss(plabels, sample_labels)
errG = errG_val + errG_label
errG.backward()
D_G_z2 = pvalidity.mean().item()
optimG.step()
print("[{}/{}] [{}/{}] D_x: [{:.4f}] D_G: [{:.4f}/{:.4f}] G_loss: [{:.4f}] D_loss: [{:.4f}] D_label: [{:.4f}] "
.format(epoch,epochs, idx, len(trainloader),D_x, D_G_z1,D_G_z2,errG,errD,
errD_real_label + errD_fake_label + errG_label))
if idx % 100 == 0:
noise = torch.randn(10,100,device = device)
labels = torch.arange(0,10,dtype = torch.long,device = device)
gen_images = gen(noise,labels).detach()
showImage(make_grid(gen_images),epoch,idx)
torch.save(gen.state_dict(),'gen.pth')
torch.save(disc.state_dict(),'disc.pth')