-
Notifications
You must be signed in to change notification settings - Fork 8
/
Train.py
160 lines (132 loc) · 6.62 KB
/
Train.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
import os
import torch
import argparse
import torch.nn as nn
import torch.optim as optim
from data import get_training_set
from libs.Matrix import MulLayer
import torchvision.utils as vutils
import torch.backends.cudnn as cudnn
from libs.utils import print_options
from libs.Criterion import LossCriterion
from libs.models import encoder4
from libs.models import decoder4
from libs.models import encoder5 as loss_network
from torch.utils.data import DataLoader
from torch.autograd import Variable
parser = argparse.ArgumentParser()
parser.add_argument("--vgg_dir", default='models/vgg_r41.pth', help='pre-trained encoder path')
parser.add_argument("--loss_network_dir", default='models/vgg_r51.pth', help='used for loss network')
parser.add_argument("--decoder_dir", default='models/dec_r41.pth', help='pre-trained decoder path')
parser.add_argument('--pretrained', type=bool, default=True)
parser.add_argument("--matrix_dir", default='model/matrix_r41_new.pth', help='pre-trained matrix path')
parser.add_argument("--data_dir", default="train_data", help='path to training dataset')
parser.add_argument('--data_augmentation', type=bool, default=True)
parser.add_argument('--threads', type=int, default=6, help='number of threads for data loader to use')
parser.add_argument("--outf", default="output/", help='folder to output images and model checkpoints')
parser.add_argument("--content_layers", default="r41", help='layers for content')
parser.add_argument("--style_layers", default="r11,r21,r31,r41", help='layers for style')
parser.add_argument("--batchSize", type=int,default=8, help='batch size')
parser.add_argument("--lr", type=float, default=1e-4, help='learning rate')
parser.add_argument("--content_weight", type=float, default=1.0, help='content loss weight')
parser.add_argument("--style_weight", type=float, default=0.02, help='style loss weight, 0.02 for origin')
parser.add_argument("--log_interval", type=int, default=500, help='log interval')
parser.add_argument("--gpu_id", type=int, default=0, help='which gpu to use')
parser.add_argument("--save_interval", type=int, default=5000, help='checkpoint save interval')
parser.add_argument("--layer", default="r41", help='which features to transfer, either r31 or r41')
parser.add_argument("--latent", type=int, default=1024, help='length of latent vector')
parser.add_argument('--nEpochs', type=int, default=5000, help='number of epochs to train for')
parser.add_argument('--snapshots', type=int, default=5, help='Snapshots')
parser.add_argument('--start_iter', type=int, default=1, help='Starting Epoch')
################# PREPARATIONS #################
opt = parser.parse_args()
opt.content_layers = opt.content_layers.split(',')
opt.style_layers = opt.style_layers.split(',')
opt.cuda = torch.cuda.is_available()
if(opt.cuda):
torch.cuda.set_device(opt.gpu_id)
os.makedirs(opt.outf,exist_ok=True)
cudnn.benchmark = True
print_options(opt)
################# DATA #################
print('===> Loading datasets')
train_set = get_training_set(opt.data_dir, opt.data_augmentation)
training_data_loader = DataLoader(dataset=train_set, num_workers=opt.threads, batch_size=opt.batchSize, shuffle=True)
################# MODEL #################
vgg5 = loss_network()
matrix = MulLayer(z_dim=opt.latent)
vgg = encoder4()
dec = decoder4()
if os.path.exists(opt.vgg_dir):
vgg.load_state_dict(torch.load(opt.vgg_dir))
print('enocder model is loaded!')
if os.path.exists(opt.decoder_dir):
dec.load_state_dict(torch.load(opt.decoder_dir))
print('decoder model is loaded!')
if os.path.exists(opt.loss_network_dir):
vgg5.load_state_dict(torch.load(opt.loss_network_dir))
print('VGG model is loaded!')
if opt.pretrained:
if os.path.exists(opt.matrix_dir):
matrix.load_state_dict(torch.load(opt.matrix_dir))
print('pretrained matrix model is loaded!')
for param in vgg.parameters():
param.requires_grad = False
for param in vgg5.parameters():
param.requires_grad = False
for param in dec.parameters():
param.requires_grad = False
################# LOSS & OPTIMIZER #################
criterion = LossCriterion(opt.style_layers,
opt.content_layers,
opt.style_weight,
opt.content_weight)
optimizer = optim.Adam(matrix.parameters(), opt.lr)
################# GPU #################
if(opt.cuda):
vgg.cuda()
dec.cuda()
vgg5.cuda()
matrix.cuda()
################# TRAINING #################
def train(epoch):
epoch_loss = 0
for iteration, batch in enumerate(training_data_loader, 1):
content, target, style = Variable(batch[0]), Variable(batch[1]), Variable(batch[2])
content = content.cuda()
target = target.cuda()
style = style.cuda()
optimizer.zero_grad()
# forward
sF = vgg(style)
cF = vgg(content)
# if(opt.layer == 'r41'):
feature, transmatrix, KL = matrix(cF[opt.layer], sF[opt.layer])
# else:
# feature, transmatrix = matrix(cF, sF)
transfer = dec(feature)
sF_loss = vgg5(style)
cF_loss = vgg5(content)
tF = vgg5(transfer)
loss, styleLoss, contentLoss, KL_loss = criterion(tF, sF_loss, cF_loss, KL)
# backward & optimization
loss.backward()
optimizer.step()
print("===> Epoch[{}]({}/{}): loss: {:.4f} || content: {:.4f} || style: {:.4f} KL: {:.4f}.".format(epoch, iteration,
len(training_data_loader), loss, contentLoss, styleLoss, KL_loss,))
print("===> Epoch {} Complete: Avg. Loss: {:.4f}".format(epoch, epoch_loss / len(training_data_loader)))
return content, style, transfer
for epoch in range(opt.start_iter, opt.nEpochs + 1):
content, style, transfer = train(epoch)
# learning rate is decayed by a factor of 10 every half of total epochs
if (epoch + 1) % 100 == 0:
for param_group in optimizer.param_groups:
param_group['lr'] /= 10.0
print('Learning rate decay: lr={}'.format(optimizer.param_groups[0]['lr']))
if epoch % (opt.snapshots) == 0:
content = content.clamp(0, 1).cpu().data
style = style.clamp(0, 1).cpu().data
transfer = transfer.clamp(0, 1).cpu().data
concat = torch.cat((content, style, transfer), dim=0)
vutils.save_image(concat, '%s/%d.png' % (opt.outf, epoch), normalize=True, scale_each=True, nrow=opt.batchSize)
torch.save(matrix.state_dict(), '%s/%s_epoch_%d.pth' % (opt.outf, opt.layer, epoch))