/
gans.py
181 lines (157 loc) · 7.94 KB
/
gans.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
import torch
import torch.nn as nn
import torch.utils.data as udata
import torch.autograd as ag
import torchvision
import models
import torch.nn.functional as F
from tqdm import tqdm, trange
import torch.utils.tensorboard
from time import time
import os
import metrics
from models import cal_grad_pen, compute_loss
class GAN(object):
def __init__(self, args, device):
self.G = models.get_g(args)
self.D = models.get_d(args)
self.G.to(device)
self.D.to(device)
self.optim_g, self.optim_d = models.get_optims(param_g=self.G.parameters(),
param_d=self.D.parameters(),
args=args)
self.z_dist = models.get_z_dist(args, device=device)
self.y_dist = models.get_y_dist(args, device=device)
self.img_size = args['data']['img_size']
self.n_channels = args['data']['n_channels']
self.loss = args['training']['loss']
self.args = args
self.device = device
self.log_dir = os.path.expanduser(self.args['training']['out_dir'] + '/' + str(time()) + '/')
os.makedirs(self.log_dir, exist_ok=True)
with open(self.log_dir + 'config.txt', 'w') as cf:
cf.write(str(self.args))
def _train_d(self, real_batch, noise_batch):
self.optim_d.zero_grad()
models.toggle_grad(self.G, False)
models.toggle_grad(self.D, True)
# train real
if len(real_batch) == 2:
real_batch_x, real_batch_y = real_batch
real_batch_x = real_batch_x.to(self.device)
real_batch_y = real_batch_y.to(self.device)
else:
real_batch_x = real_batch.to(self.device)
real_batch_y = None
pred_real = self.D(real_batch_x, real_batch_y)
loss_real = compute_loss(self.loss, pred_real, 1)
loss_real.backward()
# train fake
fake_batch = self.G(noise_batch, real_batch_y).data
pred_fake = self.D(fake_batch, real_batch_y)
loss_fake = compute_loss(self.loss, pred_fake, 0)
loss_fake.backward()
# grad pen
grad_pen = cal_grad_pen(self.D, real_batch=real_batch_x, fake_batch=fake_batch,
gp_weight=self.args['training']['gp_weight'],
gp_inter=self.args['training']['gp_inter'],
gp_center=self.args['training']['gp_center'],
label_batch=real_batch_y)
grad_pen.backward()
# update params
self.optim_d.step()
loss_d = loss_real + loss_fake + grad_pen
return loss_d.item()
def _train_g(self, noise_batch, label_batch):
self.optim_g.zero_grad()
models.toggle_grad(self.G, True)
models.toggle_grad(self.D, False)
fake_batch = self.G(noise_batch, label_batch)
pred_fake = self.D(fake_batch, label_batch)
loss_fake = compute_loss(self.loss, pred_fake, 1)
loss_fake.backward()
# update params
self.optim_g.step()
return loss_fake.item()
def train(self):
train_data, test_data, n_labels = models.load_data(self.args)
n_epochs = self.args['training']['n_epochs']
batch_size = self.args['training']['batch_size']
d_steps = self.args['training']['d_steps']
log_interval = self.args['training']['log_interval']
self.writer = torch.utils.tensorboard.SummaryWriter(log_dir=self.log_dir + '/run/')
self.fixed_z = self.z_dist.sample((64,))
self.fixed_y = self.y_dist.sample((64,))
self.z_start = self.z_dist.sample((32,))
self.z_end = self.z_dist.sample((32,))
self.z_inter_list = metrics.slerp(self.z_start, self.z_end, 14)
for eidx in trange(n_epochs, leave=True, desc='Epoch'):
for iidx, real_batch in enumerate(tqdm(train_data)):
noise_batch = self.z_dist.sample((batch_size,))
loss_d = self._train_d(real_batch=real_batch, noise_batch=noise_batch)
self.writer.add_scalar('Loss/%d/d' % eidx, loss_d, global_step=iidx)
if iidx % d_steps == 0:
# train G
if len(real_batch) == 2:
label_batch = real_batch[1]
else:
label_batch = None
loss_g = self._train_g(noise_batch=noise_batch, label_batch=label_batch)
self.writer.add_scalar('Loss/%d/g' % eidx, loss_g, global_step=iidx)
if eidx % log_interval == 0:
self._log(eidx, test_data)
def _evaluate(self):
pass
def _log(self, eidx, test_data):
img_size = self.args['data']['img_size']
n_channels = self.args['data']['n_channels']
with torch.no_grad():
fixed_fake = self.G(self.fixed_z, self.fixed_y)
fixed_fake = fixed_fake.view(fixed_fake.size(0), self.n_channels, self.img_size, self.img_size)
torchvision.utils.save_image(fixed_fake, self.log_dir + 'fake_%05d.jpg' % eidx)
inter_fake = [self.G(z) for z in self.z_inter_list]
inter_fake = torch.cat(inter_fake, dim=0)
inter_fake = inter_fake.view(inter_fake.size(0), -1,
self.args['data']['img_size'],
self.args['data']['img_size'])
torchvision.utils.save_image(inter_fake, self.log_dir + 'inter_%05d.jpg' % eidx,
nrow=32, normalize=inter_fake.size(1) > 1)
if self.args['nnd']['enable']:
print('computing nnd')
# generate data
sample_size = self.args['nnd']['sample_size']
batch_size = self.args['training']['batch_size']
n_batches = sample_size // batch_size + 1
fake_data = []
with torch.no_grad():
for _ in range(n_batches):
z = self.z_dist.sample((batch_size, ))
y = self.y_dist.sample((batch_size, ))
fake_batch = self.G(z, y)
fake_data.append(fake_batch)
fake_data = torch.cat(fake_data, dim=0)
fake_data = udata.TensorDataset(fake_data)
fake_data = udata.DataLoader(fake_data, batch_size=batch_size, shuffle=True, drop_last=True)
# compute 2 nnd scores
# build nets
net = models.get_c(self.args)
nnd_fixed = metrics.nnd_iter(C=net, gan_loss='wgan', real_data=test_data,
fake_data=fake_data, lr=self.args['nnd']['lr'], betas=(0.9, 0.999),
noise_weight=self.args['nnd']['noise_weight'],
noise_dist=self.args['nnd']['noise_dist'],
gp_weight=self.args['nnd']['gp_weight'],
n_iter=self.args['nnd']['n_iters'], device=self.device)
# build nets
net = models.get_c(self.args)
nnd_inf = metrics.nnd_iter_gen(C=net, G=self.G, gan_loss='wgan', real_data=test_data,
z_dist=self.z_dist, y_dist=self.y_dist, lr=self.args['nnd']['lr'],
betas=(0.9, 0.999), noise_weight=self.args['nnd']['noise_weight'],
noise_dist=self.args['nnd']['noise_dist'],
gp_weight=self.args['nnd']['gp_weight'],
n_iter=self.args['nnd']['n_iters'], device=self.device)
# log
self.writer.add_scalar('nnd_fixed_%05d' % eidx, nnd_fixed, global_step=eidx)
self.writer.add_scalar('nnd_inf_%05d' % eidx, nnd_inf, global_step=eidx)
with open(self.log_dir + 'nnd.txt', 'a+') as nnd_file:
nnd_file.write('%d %f %f\n' % (eidx, nnd_fixed, nnd_inf))
# end if