-
Notifications
You must be signed in to change notification settings - Fork 12
/
single_gan.py
235 lines (215 loc) · 10.7 KB
/
single_gan.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
from __future__ import print_function
from collections import OrderedDict
import torch
import torch.nn as nn
import torch.optim as optim
import torch.backends.cudnn as cudnn
from torch.autograd import Variable
from torchvision.utils import make_grid
from models.model import D_NET_Multi, SingleGenerator, Encoder, weights_init
from util.loss import GANLoss, KL_loss
from util.util import tensor2im
import numpy as np
################## SingleGAN #############################
class SingleGAN():
def name(self):
return 'SingleGAN'
def initialize(self, opt):
torch.cuda.set_device(opt.gpu)
cudnn.benchmark = True
self.opt = opt
self.build_models()
def build_models(self):
################### generator #########################################
self.G = SingleGenerator(input_nc=self.opt.input_nc, output_nc=self.opt.input_nc, ngf=self.opt.ngf, nc=self.opt.c_num+self.opt.d_num, e_blocks=self.opt.e_blocks, norm_type=self.opt.norm)
################### encoder ###########################################
self.E =None
if self.opt.mode == 'multimodal':
self.E = Encoder(input_nc=self.opt.input_nc, output_nc=self.opt.c_num, nef=self.opt.nef, nd=self.opt.d_num, n_blocks=4, norm_type=self.opt.norm)
if self.opt.isTrain:
################### discriminators #####################################
self.Ds = []
for i in range(self.opt.d_num):
self.Ds.append(D_NET_Multi(input_nc=self.opt.output_nc, ndf=self.opt.ndf, block_num=3,norm_type=self.opt.norm))
################### init_weights ########################################
if self.opt.continue_train:
self.G.load_state_dict(torch.load('{}/G_{}.pth'.format(self.opt.model_dir, self.opt.which_epoch)))
if self.E is not None:
self.E.load_state_dict(torch.load('{}/E_{}.pth'.format(self.opt.model_dir, self.opt.which_epoch)))
for i in range(self.opt.d_num):
self.Ds[i].load_state_dict(torch.load('{}/D_{}_{}.pth'.format(self.opt.model_dir, i, self.opt.which_epoch)))
else:
self.G.apply(weights_init(self.opt.init_type))
if self.E is not None:
self.E.apply(weights_init(self.opt.init_type))
for i in range(self.opt.d_num):
self.Ds[i].apply(weights_init(self.opt.init_type))
################### use GPU #############################################
self.G.cuda()
if self.E is not None:
self.E.cuda()
for i in range(self.opt.d_num):
self.Ds[i].cuda()
################### set criterion ########################################
self.criterionGAN = GANLoss(mse_loss=(self.opt.c_gan_mode == 'lsgan'))
################## define optimizers #####################################
self.define_optimizers()
else:
self.G.load_state_dict(torch.load('{}/G_{}.pth'.format(self.opt.model_dir, self.opt.which_epoch)))
self.G.cuda()
self.G.eval()
if self.E is not None:
self.E.load_state_dict(torch.load('{}/E_{}.pth'.format(self.opt.model_dir, self.opt.which_epoch)))
self.E.cuda()
self.E.eval()
def sample_latent_code(self, size):
c = torch.cuda.FloatTensor(size).normal_()
return Variable(c)
def get_domain_code(self, domainLable):
domainCode = torch.zeros([len(domainLable),self.opt.d_num])
domainIndex_cache = [[] for i in range(self.opt.d_num)]
for index in range(len(domainLable)):
domainCode[index, domainLable[index]] = 1
domainIndex_cache[domainLable[index]].append(index)
domainIndex = []
for index in domainIndex_cache:
domainIndex.append(Variable(torch.LongTensor(index)).cuda())
return Variable(domainCode).cuda(), domainIndex
def define_optimizer(self, Net):
return optim.Adam(Net.parameters(),
lr=self.opt.lr,
betas=(0.5, 0.999))
def define_optimizers(self):
self.G_opt = self.define_optimizer(self.G)
self.E_opt = None
if self.E is not None:
self.E_opt = self.define_optimizer(self.E)
self.Ds_opt = []
for i in range(self.opt.d_num):
self.Ds_opt.append(self.define_optimizer(self.Ds[i]))
def update_lr(self, lr):
for param_group in self.G_opt.param_groups:
param_group['lr'] = lr
if self.E_opt is not None:
for param_group in self.E_opt.param_groups:
param_group['lr'] = lr
for i in range(self.opt.d_num):
for param_group in self.Ds_opt[i].param_groups:
param_group['lr'] = lr
def save(self, name):
torch.save(self.G.state_dict(), '{}/G_{}.pth'.format(self.opt.model_dir, name))
if self.E_opt is not None:
torch.save(self.E.state_dict(), '{}/E_{}.pth'.format(self.opt.model_dir, name))
for i in range(self.opt.d_num):
torch.save(self.Ds[i].state_dict(), '{}/D_{}_{}.pth'.format(self.opt.model_dir, i, name))
def prepare_image(self, data):
img, sourceD, targetD = data
return Variable(torch.cat(img,0)).cuda(), torch.cat(sourceD,0), torch.cat(targetD,0)
def translation(self, data):
input, sourceD, targetD = self.prepare_image(data)
sourceDC, sourceIndex = self.get_domain_code(sourceD)
targetDC, targetIndex = self.get_domain_code(targetD)
images, names =[], []
for i in range(self.opt.d_num):
images.append([tensor2im(input.index_select(0,sourceIndex[i])[0].data)])
names.append(['D{}'.format(i)])
if self.opt.mode == 'multimodal':
for i in range(self.opt.n_samples):
c_rand = self.sample_latent_code(torch.Size([input.size(0),self.opt.c_num]))
targetC = torch.cat([targetDC, c_rand],1)
output = self.G(input,targetC)
for j in range(output.size(0)):
images[sourceD[j]].append(tensor2im(output[j].data))
names[sourceD[j]].append('{}to{}_{}'.format(sourceD[j],targetD[j],i))
else:
output = self.G(input,targetDC)
for i in range(output.size(0)):
images[sourceD[i]].append(tensor2im(output[i].data))
names[sourceD[i]].append('{}to{}'.format(sourceD[i],targetD[i]))
return images, names
def get_current_errors(self):
dict = []
for i in range(self.opt.d_num):
dict += [('D_{}'.format(i), self.errDs[i].data[0])]
dict += [('G_{}'.format(i), self.errGs[i].data[0])]
dict += [('errCyc', self.errCyc.data[0])]
if self.opt.lambda_ide > 0:
dict += [('errIde', self.errIde.data[0])]
if self.E is not None:
dict += [('errKl', self.errKL.data[0])]
dict += [('errCode', self.errCode.data[0])]
return OrderedDict(dict)
def get_current_visuals(self):
real = make_grid(self.real.data,nrow=self.real.size(0),padding=0)
fake = make_grid(self.fake.data,nrow=self.real.size(0),padding=0)
cyc = make_grid(self.cyc.data,nrow=self.real.size(0),padding=0)
img = [real,fake,cyc]
name = 'rsal,fake,cyc'
if self.opt.lambda_ide > 0:
ide = make_grid(self.ide.data,nrow=self.real.size(0),padding=0)
img.append(ide)
name +=',ide'
img = torch.cat(img,1)
return OrderedDict([(name,tensor2im(img))])
def update_D(self, D, D_opt, real, fake):
D.zero_grad()
pred_fake = D(fake.detach())
pred_real = D(real)
errD = self.criterionGAN(pred_fake,False) + self.criterionGAN(pred_real,True)
errD.backward()
D_opt.step()
return errD
def calculate_G(self, D, fake):
pred_fake = D(fake)
errG = self.criterionGAN(pred_fake,True)
return errG
def update_model(self,data):
### prepare data ###
self.real, sourceD, targetD = self.prepare_image(data)
sourceDC, self.sourceIndex = self.get_domain_code(sourceD)
targetDC, self.targetIndex = self.get_domain_code(targetD)
sourceC, targetC = sourceDC, targetDC
### generate image ###
if self.E is not None:
c_enc, mu, logvar = self.E(self.real,sourceDC)
c_rand = self.sample_latent_code(c_enc.size())
sourceC = torch.cat([sourceDC, c_enc],1)
targetC = torch.cat([targetDC, c_rand],1)
self.fake = self.G(self.real,targetC)
self.cyc = self.G(self.fake,sourceC)
if self.E is not None:
_, mu_enc, _ = self.E(self.fake,targetDC)
if self.opt.lambda_ide > 0:
self.ide = self.G(self.real,sourceC)
### update D ###
self.errDs = []
for i in range(self.opt.d_num):
errD = self.update_D(self.Ds[i], self.Ds_opt[i], self.real.index_select(0,self.sourceIndex[i]), self.fake.index_select(0,self.targetIndex[i]))
self.errDs.append(errD)
### update G ###
self.errGs, self.errKl, self.errCode, errG_total = [], 0, 0, 0
self.G.zero_grad()
for i in range(self.opt.d_num):
errG = self.calculate_G(self.Ds[i], self.fake.index_select(0,self.targetIndex[i]))
errG_total += errG
self.errGs.append(errG)
self.errCyc = torch.mean(torch.abs(self.cyc-self.real)) * self.opt.lambda_cyc
errG_total += self.errCyc
if self.opt.lambda_ide > 0:
self.errIde = torch.mean(torch.abs(self.ide-self.real)) * self.opt.lambda_ide
errG_total += self.errIde
if self.E is not None:
self.E.zero_grad()
self.errKL = KL_loss(mu,logvar) * self.opt.lambda_kl
errG_total += self.errKL
errG_total.backward(retain_graph=True)
self.G_opt.step()
self.E_opt.step()
self.G.zero_grad()
self.E.zero_grad()
self.errCode = torch.mean(torch.abs(mu_enc - c_rand)) * self.opt.lambda_c
self.errCode.backward()
self.G_opt.step()
else:
errG_total.backward()
self.G_opt.step()