-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrainer.py
247 lines (206 loc) · 10.3 KB
/
trainer.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
######################################################################
# script name : trainer.py
# author : Chen Xuanhong
# created time : 2019/9/11 22:36
# modification time ：2019/9/13 11:35
# modified by : Chen Xuanhong
######################################################################
import os
import time
import datetime
import functools
import torch
import torch.nn as nn
from torch.autograd import Variable
from torchvision.utils import save_image
from utilities.Utilities import *
from tensorboardX import SummaryWriter
from utilities.Reporter import Reporter
import utilities.Sampler as Sampler
from torchviz import make_dot
from components.GenResNet32 import ResNetGenerator
from components.SNResNetProjectionDiscriminator32 import SNResNetProjectionDiscriminator
import metrics.FID as FIDCaculator
#from inceptionScoreMetricClass import inceptionScoreMetricClass
class Trainer(object):
def __init__(self, data_loader, config):
self.report_file = os.path.join(config.log_path, config.version,config.version+"_report.log")
self.reporter = Reporter(self.report_file)
# Data loader
self.data_loader = data_loader
# exact model and loss
self.cGAN = config.cGAN
self.adv_loss = config.adv_loss
# Model hyper-parameters
self.imsize = config.imsize
self.z_dim = config.z_dim
self.g_conv_dim = config.g_conv_dim
self.d_conv_dim = config.d_conv_dim
self.n_classes = config.n_class if config.cGAN else 0
self.parallel = config.parallel
self.seed = config.seed
self.device = torch.device('cuda:%d'%config.cuda)
self.GPUs = config.GPUs
self.gen_distribution = config.gen_distribution
self.gen_bottom_width = config.gen_bottom_width
self.total_step = config.total_step
self.batch_size = config.batch_size
self.num_workers= config.num_workers
self.g_lr = config.g_lr
self.d_lr = config.d_lr
self.lr_decay = config.lr_decay
self.beta1 = config.beta1
self.beta2 = config.beta2
self.use_pretrained_model = config.use_pretrained_model
self.chechpoint_step = config.chechpoint_step
self.use_pretrained_model = config.use_pretrained_model
self.dataset = config.dataset
self.image_path = config.image_path
self.log_path = config.log_path
self.model_save_path= config.model_save_path
self.sample_path = config.sample_path
self.log_step = config.log_step
self.sample_step = config.sample_step
self.model_save_step= config.model_save_step
self.version = config.version
self.caculate_FID = config.caculate_FID
self.DStep = config.D_step
self.GStep = config.G_step
self.metric_caculation_step = config.metric_caculation_step
# Path
self.log_path = os.path.join(config.log_path, self.version)
self.summary_path = self.log_path
self.sample_path = os.path.join(config.sample_path, self.version)
self.model_save_path= os.path.join(config.model_save_path, self.version)
self.build_model()
self.reporter.writeConfig(config)
self.reporter.writeModel(self.G.__str__())
self.reporter.writeModel(self.D.__str__())
if self.caculate_FID:
z_sampler,c_sampler = Sampler.prepare_z_c(self.batch_size,self.z_dim,self.n_classes,device=self.device)
gsampler = functools.partial(Sampler.sampleG,G=self.G,z_=z_sampler,c_=c_sampler, parallel=self.parallel)
self.get_inception_metrics = FIDCaculator.prepare_inception_metrics(config.FID_mean_cov,gsampler,config.metric_images_num)
self.writer = SummaryWriter(log_dir=self.summary_path)
z = torch.zeros(1, self.z_dim).to(self.device)
c = torch.zeros(1).long().to(self.device)
y = torch.zeros(1,3,self.imsize,self.imsize).to(self.device)
vise_graph = make_dot(self.G(z,c), params=dict(self.G.named_parameters()))
vise_graph.view(self.log_path+"/Generator")
vise_graph = make_dot(self.D(y,c), params=dict(self.D.named_parameters()))
vise_graph.view(self.log_path+"/Discriminator",quiet=False,quiet_view=False)
del z
del c
del y
# Start with trained model
if self.use_pretrained_model:
self.load_pretrained_model()
def build_model(self):
self.G = ResNetGenerator(self.g_conv_dim, self.z_dim, self.gen_bottom_width,
num_classes=self.n_classes).to(self.device)
self.D = SNResNetProjectionDiscriminator(self.d_conv_dim, self.n_classes).to(self.device)
if self.parallel:
self.G = nn.DataParallel(self.G,device_ids=self.GPUs)
self.D = nn.DataParallel(self.D,device_ids=self.GPUs)
# Loss and optimizer
self.g_optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, self.G.parameters()), self.g_lr, [self.beta1, self.beta2])
self.d_optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, self.D.parameters()), self.d_lr, [self.beta1, self.beta2])
def train(self):
# Data iterator
data_iter = iter(self.data_loader)
model_save_step = self.model_save_step
# Fixed input for debugging
sampleBatch = 10
fixed_z = torch.randn(self.n_classes*sampleBatch, self.z_dim)
fixed_z = fixed_z.to(self.device)
fixed_c = Sampler.sampleFixedLabels(self.n_classes,sampleBatch,self.device)
runingZ,runingLabel = Sampler.prepare_z_c(self.batch_size, self.z_dim, self.n_classes, device=self.device)
# Start with trained model
if self.use_pretrained_model:
start = self.chechpoint_step + 1
else:
start = 0
# Start time
start_time = time.time()
self.reporter.writeInfo("Start to train the model")
dstepCounter= 0
gstepCounter= 0
for step in range(start, self.total_step):
# ================== Train D ================== #
self.D.train()
self.G.train()
if dstepCounter<self.DStep:
try:
realImages, realLabel = next(data_iter)
except:
data_iter = iter(self.data_loader)
realImages, realLabel = next(data_iter)
# Compute loss with real images
realImages = realImages.to(self.device)
realLabel = realLabel.to(self.device).long()
d_out_real = self.D(realImages,realLabel)
d_loss_real = torch.nn.ReLU()(1.0 - d_out_real).mean()
# apply Gumbel Softmax
runingZ.sample_()
runingLabel.sample_()
fake_images = self.G(runingZ,runingLabel)
d_out_fake = self.D(fake_images,runingLabel)
d_loss_fake = torch.nn.ReLU()(1.0 + d_out_fake).mean()
# Backward + Optimize
d_loss = d_loss_real + d_loss_fake
self.reset_grad()
d_loss.backward()
self.d_optimizer.step()
dstepCounter += 1
else:
# ================== Train G and gumbel ================== #
# Create random noise
runingZ.sample_()
runingLabel.sample_()
fake_images = self.G(runingZ,runingLabel)
# Compute loss with fake images
g_out_fake = self.D(fake_images,runingLabel)
g_loss_fake = - g_out_fake.mean()
self.reset_grad()
g_loss_fake.backward()
self.g_optimizer.step()
gstepCounter += 1
if gstepCounter==self.GStep:
dstepCounter = 0
gstepCounter = 0
# Print out log info
if (step + 1) % self.log_step == 0:
elapsed = time.time() - start_time
elapsed = str(datetime.timedelta(seconds=elapsed))
print("Elapsed [{}], G_step [{}/{}], D_step[{}/{}], d_out_real: {:.4f}, "
" d_loss_fake: {:.4f}, g_loss_fake: {:.4f}".
format(elapsed, step + 1, self.total_step, (step + 1),
self.total_step , d_loss_real.item(),
d_loss_fake.item(), g_loss_fake.item()))
self.writer.add_scalar('log/d_loss_real', d_loss_real.item(),(step + 1))
self.writer.add_scalar('log/d_loss_fake', d_loss_fake.item(),(step + 1))
self.writer.add_scalar('log/d_loss', d_loss.item(), (step + 1))
self.writer.add_scalar('log/g_loss_fake', g_loss_fake.item(), (step + 1))
if (step + 1) % self.sample_step == 0:
fake_images = self.G(fixed_z,fixed_c)
save_image(denorm(fake_images.data),
os.path.join(self.sample_path, '{}_fake.png'.format(step + 1)),nrow=self.n_classes)
if (step+1) % model_save_step==0:
torch.save(self.G.state_dict(),
os.path.join(self.model_save_path, '{}_G.pth'.format(step + 1)))
torch.save(self.D.state_dict(),
os.path.join(self.model_save_path, '{}_D.pth'.format(step + 1)))
if (step+1) % self.metric_caculation_step == 0 and self.caculate_FID:
print("start to caculate the FID")
FID = self.get_inception_metrics()
print("FID is %.3f"%FID)
self.writer.add_scalar('metric/FID', FID, (step + 1))
self.reporter.writeTrainLog(step+1,"Current FID is %.4f"%FID)
def load_pretrained_model(self):
self.G.load_state_dict(torch.load(os.path.join(
self.model_save_path, '{}_G.pth'.format(self.chechpoint_step))))
self.D.load_state_dict(torch.load(os.path.join(
self.model_save_path, '{}_D.pth'.format(self.chechpoint_step))))
print('loaded trained models (step: {})..!'.format(self.chechpoint_step))
def reset_grad(self):
self.d_optimizer.zero_grad()
self.g_optimizer.zero_grad()