Newer
Older
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
"""solver.py"""
import time
from pathlib import Path
import visdom
import torch
import torch.optim as optim
import torch.nn.functional as F
from torch.autograd import Variable
from torchvision.utils import make_grid
from utils import cuda
from model import BetaVAE
from dataset import return_data
def original_vae_loss(x, x_recon, mu, logvar):
batch_size = x.size(0)
if batch_size == 0:
recon_loss = 0
kl_divergence = 0
else:
recon_loss = F.binary_cross_entropy_with_logits(x_recon, x, size_average=False).div(batch_size)
# kld which one is correct? from the equation in most of papers,
# I think the first one is correct but official pytorch code uses the second one.
kl_divergence = -0.5*(1 + logvar - mu**2 - logvar.exp()).sum(1).mean()
#kl_divergence = -0.5*(1 + logvar - mu**2 - logvar.exp()).sum()
#dimension_wise_kl_divergence = -0.5*(1 + logvar - mu**2 - logvar.exp()).mean(0)
return recon_loss, kl_divergence
class Solver(object):
def __init__(self, args):
# Misc
self.use_cuda = args.cuda and torch.cuda.is_available()
self.max_iter = args.max_iter
self.global_iter = 0
# Networks & Optimizers
self.z_dim = args.z_dim
self.beta = args.beta
self.lr = args.lr
self.beta1 = args.beta1
self.beta2 = args.beta2
self.net = cuda(BetaVAE(self.z_dim), self.use_cuda)
self.optim = optim.Adam(self.net.parameters(), lr=self.lr,
betas=(self.beta1, self.beta2))
# Visdom
self.viz_name = args.viz_name
self.viz_port = args.viz_port
self.viz_on = args.viz_on
if self.viz_on:
self.viz = visdom.Visdom(env=self.viz_name, port=self.viz_port)
self.viz_curves = visdom.Visdom(env=self.viz_name+'/train_curves', port=self.viz_port)
self.win_recon = None
self.win_kld = None
# Checkpoint
self.ckpt_dir = Path(args.ckpt_dir).joinpath(args.viz_name)
if not self.ckpt_dir.exists():
self.ckpt_dir.mkdir(parents=True, exist_ok=True)
self.load_ckpt = args.load_ckpt
if self.load_ckpt:
self.load_checkpoint()
# Data
self.dset_dir = args.dset_dir
self.batch_size = args.batch_size
self.data_loader = return_data(args)
def train(self):
self.net_mode(train=True)
out = False
while not out:
start = time.time()
curve_data = []
for x in self.data_loader:
self.global_iter += 1
x = Variable(cuda(x, self.use_cuda))
x_recon, mu, logvar = self.net(x)
recon_loss, kld = original_vae_loss(x, x_recon, mu, logvar)
beta_vae_loss = recon_loss + self.beta*kld
self.optim.zero_grad()
beta_vae_loss.backward()
self.optim.step()
if self.global_iter%1000 == 0:
curve_data.append(torch.Tensor([self.global_iter,
recon_loss.data[0],
kld.data[0],]))
if self.global_iter%5000 == 0:
self.save_checkpoint()
self.visualize(dict(image=[x, x_recon], curve=curve_data))
print('[{}] recon_loss:{:.3f} beta*kld:{:.3f}'.format(
self.global_iter, recon_loss.data[0], self.beta*kld.data[0]))
curve_data = []
if self.global_iter >= self.max_iter:
out = True
break
end = time.time()
print('[time elapsed] {:.2f}s/epoch'.format(end-start))
print("[Training Finished]")
def visualize(self, data):
x, x_recon = data['image']
curve_data = data['curve']
sample_x = make_grid(x.data.cpu(), normalize=False)
sample_x_recon = make_grid(F.sigmoid(x_recon).data.cpu(), normalize=False)
samples = torch.stack([sample_x, sample_x_recon], dim=0)
self.viz.images(samples, opts=dict(title=str(self.global_iter)))
curve_data = torch.stack(curve_data, dim=0)
curve_iter = curve_data[:, 0]
curve_recon = curve_data[:, 1]
curve_kld = curve_data[:, 2]
if self.win_recon is None:
self.win_recon = self.viz_curves.line(
X=curve_iter,
Y=curve_recon,
opts=dict(
xlabel='iteration',
ylabel='reconsturction loss',))
else:
self.win_recon = self.viz_curves.line(
X=curve_iter,
Y=curve_recon,
win=self.win_recon,
update='append',
opts=dict(
xlabel='iteration',
ylabel='reconsturction loss',))
if self.win_kld is None:
self.win_kld = self.viz_curves.line(
X=curve_iter,
Y=curve_kld,
opts=dict(
xlabel='iteration',
ylabel='kl divergence',))
else:
self.win_kld = self.viz_curves.line(
X=curve_iter,
Y=curve_kld,
win=self.win_kld,
update='append',
opts=dict(
xlabel='iteration',
ylabel='kl divergence',))
def traverse(self):
import random
decoder = self.net.decode
encoder = self.net.encode
viz = visdom.Visdom(env=self.viz_name+'/traverse', port=self.viz_port)
n_dsets = self.data_loader.dataset.__len__()
fixed_idx = 0
rand_idx = random.randint(1, n_dsets-1)
fixed_img = self.data_loader.dataset.__getitem__(fixed_idx)
fixed_img = Variable(cuda(fixed_img, self.use_cuda), volatile=True).unsqueeze(0)
fixed_img_z = encoder(fixed_img)[:, :self.z_dim]
random_img = self.data_loader.dataset.__getitem__(rand_idx)
random_img = Variable(cuda(random_img, self.use_cuda), volatile=True).unsqueeze(0)
zero_z = Variable(cuda(torch.zeros(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
random_z = Variable(cuda(torch.rand(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
Z = {'fixed_img':fixed_img_z, 'random_img':random_img_z, 'random_z':random_z, 'zero_z':zero_z}
for key in Z.keys():
z_ori = Z[key]
samples = []
for row in range(self.z_dim):
z = z_ori.clone()
for val in interpolation:
z[:, row] = val
sample = F.sigmoid(decoder(z))
samples.append(sample)
samples = torch.cat(samples, dim=0).data.cpu()
viz.images(samples, opts=dict(title=title), nrow=len(interpolation))
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
def net_mode(self, train):
if not isinstance(train, bool):
raise('Only bool type is supported. True or False')
if train:
self.net.train()
else:
self.net.eval()
def save_checkpoint(self, filename='ckpt.tar', silent=True):
model_states = {'net':self.net.state_dict(),}
optim_states = {'optim':self.optim.state_dict(),}
win_states = {'recon':self.win_recon,
'kld':self.win_kld,}
states = {'iter':self.global_iter,
'win_states':win_states,
'model_states':model_states,
'optim_states':optim_states}
file_path = self.ckpt_dir.joinpath(filename)
torch.save(states, file_path.open('wb+'))
if not silent:
print("=> saved checkpoint '{}' (iter {})".format(file_path, self.global_iter))
def load_checkpoint(self, filename='ckpt.tar'):
file_path = self.ckpt_dir.joinpath(filename)
if file_path.is_file():
checkpoint = torch.load(file_path.open('rb'))
self.global_iter = checkpoint['iter']
self.win_recon = checkpoint['win_states']['recon']
self.win_kld = checkpoint['win_states']['kld']
self.net.load_state_dict(checkpoint['model_states']['net'])
self.optim.load_state_dict(checkpoint['optim_states']['optim'])
print("=> loaded checkpoint '{} (iter {})'".format(file_path, self.global_iter))
else:
print("=> no checkpoint found at '{}'".format(file_path))