-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathupdater.py
129 lines (103 loc) · 4.2 KB
/
updater.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
import numpy as np
import os, sys
import chainer
import chainer.functions as F
from chainer import Variable
sys.path.append(os.path.dirname(__file__))
sys.path.append(os.path.abspath(os.path.dirname(__file__)) + os.path.sep + os.path.pardir)
from common.misc import soft_copy_param,copy_param,average_param
# Classic Adversarial Loss
def loss_dcgan_dis(dis_fake, dis_real):
L1 = F.mean(F.softplus(-dis_real))
L2 = F.mean(F.softplus(dis_fake))
loss = L1 + L2
return loss
def loss_dcgan_gen(dis_fake):
loss = F.mean(F.softplus(-dis_fake))
return loss
# Hinge Loss
def loss_hinge_dis(dis_fake, dis_real):
loss = F.mean(F.relu(1. - dis_real))
loss += F.mean(F.relu(1. + dis_fake))
return loss
def loss_hinge_gen(dis_fake):
loss = -F.mean(dis_fake)
return loss
# WGAN loss
def loss_wgan_dis(dis_fake, dis_real):
L1 = F.sum(-dis_real)
L2 = F.sum(dis_fake)
loss = L1 + L2
return loss
def loss_wgan_gen(dis_fake):
loss = F.sum(-dis_fake)
return loss
class Updater(chainer.training.StandardUpdater):
def __init__(self, *args, **kwargs):
self.gen, self.dis, self.g_ema,self.g_ma = kwargs.pop('models')
self.n_dis = kwargs.pop('n_dis')
self.smoothing = kwargs.pop('smoothing')
self.objective = kwargs.pop('objective')
self.ma_start = kwargs.pop('ma_start')
self.counter = 0
self.n_model = 0
self.lam =10.0
if self.objective == 'gan':
self.loss_gen = loss_dcgan_gen
self.loss_dis = loss_dcgan_dis
elif self.objective == 'hinge':
self.loss_gen = loss_hinge_gen
self.loss_dis = loss_hinge_dis
elif self.objective == 'wgan-gp':
self.loss_gen = loss_wgan_gen
self.loss_dis = loss_wgan_dis
else:
NotImplementedError('no such objective')
super(Updater, self).__init__(*args, **kwargs)
def update_core(self):
gen_optimizer = self.get_optimizer('opt_gen')
dis_optimizer = self.get_optimizer('opt_dis')
xp = self.gen.xp
self.counter += 1
for i in range(self.n_dis):
batch = self.get_iterator('main').next()
batchsize = len(batch)
x = []
for j in range(batchsize):
x.append(np.asarray(batch[j]).astype("f"))
x_real = Variable(xp.asarray(x))
if i == 0:
# train generator
z = Variable(xp.asarray(self.gen.make_hidden(batchsize)))
x_fake = self.gen(z)
y_fake = self.dis(x_fake)
loss_gen = self.loss_gen(y_fake)#F.sum(F.softplus(-y_fake)) / batchsize
self.gen.cleargrads()
loss_gen.backward()
gen_optimizer.update()
chainer.reporter.report({'loss_gen': loss_gen})
soft_copy_param(self.g_ema, self.gen, 1.0-self.smoothing)
if self.counter >= self.ma_start:
average_param(self.g_ma,self.gen,self.n_model)
self.n_model += 1
y_real = self.dis(x_real)
z = Variable(xp.asarray(self.gen.make_hidden(batchsize)))
x_fake = self.gen(z)
y_fake = self.dis(x_fake)
x_fake.unchain_backward()
loss_dis = self.loss_dis(y_fake,y_real)
if self.objective =='wgan-gp':
eps = xp.random.uniform(0, 1, size=batchsize).astype("f")[:, None, None, None]
x_mid = eps * x_real + (1.0 - eps) * x_fake
x_mid_v = Variable(x_mid.data)
y_mid = F.sum(self.dis(x_mid_v))
dydx, = chainer.grad([y_mid], [x_mid_v], enable_double_backprop=True)
dydx = F.sqrt(F.sum((dydx*dydx), axis=(1, 2, 3)))
loss_gp = self.lam * F.mean_squared_error(dydx, xp.ones_like(dydx.data))
loss_dis += loss_gp
chainer.reporter.report({'loss_gp': loss_gp})
self.dis.cleargrads()
loss_dis.backward()
dis_optimizer.update()
loss_dis.unchain_backward()
chainer.reporter.report({'loss_dis': loss_dis})