forked from Aaditya-Singh/SAFIN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
200 lines (165 loc) · 7.63 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
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
import argparse
import random
from importlib import import_module
from pathlib import Path
import os
import numpy as np
import torch
import torch.backends.cudnn as cudnn
import torch.utils.data as data
from PIL import Image, ImageFile
from tensorboardX import SummaryWriter
from torchvision import transforms
from torchvision.utils import make_grid
from tqdm import tqdm
from sampler import InfiniteSamplerWrapper
from test import style_transfer
cudnn.benchmark = True
Image.MAX_IMAGE_PIXELS = None # Disable DecompressionBombError
ImageFile.LOAD_TRUNCATED_IMAGES = True # Disable OSError: image file is truncated
def prepare_seed(rand_seed):
random.seed(rand_seed)
np.random.seed(rand_seed)
torch.manual_seed(rand_seed)
torch.cuda.manual_seed(rand_seed)
torch.cuda.manual_seed_all(rand_seed)
def train_transform():
transform_list = [
transforms.Resize(size=(512, 512)),
transforms.RandomCrop(256),
transforms.ToTensor()
]
return transforms.Compose(transform_list)
class FlatFolderDataset(data.Dataset):
def __init__(self, root, transform):
super(FlatFolderDataset, self).__init__()
self.root = root
self.paths = list(Path(self.root).glob('*'))
self.transform = transform
def __getitem__(self, index):
path = self.paths[index]
img = Image.open(str(path)).convert('RGB')
img = self.transform(img)
return img
def __len__(self):
return len(self.paths)
def name(self):
return 'FlatFolderDataset'
def adjust_learning_rate(optimizer, iteration_count):
"""Imitating the original implementation"""
lr = args.lr / (1.0 + args.lr_decay * iteration_count)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
if __name__ == '__main__':
parser = argparse.ArgumentParser()
# Basic options
parser.add_argument('--content_dir', type=str, required=True,
help='Directory path to a batch of content images')
parser.add_argument('--style_dir', type=str, required=True,
help='Directory path to a batch of style images')
parser.add_argument('--vgg', type=str, default='models/vgg_normalised.pth')
# training options
parser.add_argument('--save_dir',required=True,
help='Directory to save the model')
parser.add_argument('--log_dir', default='./logs',
help='Directory to save the log')
parser.add_argument('--lr', type=float, default=1e-4)
parser.add_argument('--lr_decay', type=float, default=5e-5)
parser.add_argument('--max_iter', type=int, default=160000)
parser.add_argument('--batch_size', type=int, default=8)
parser.add_argument('--style_weight', type=float, default=10.0)
parser.add_argument('--content_weight', type=float, default=1.0)
parser.add_argument('--n_threads', type=int, default=16)
parser.add_argument('--save_model_interval', type=int, default=10000)
parser.add_argument('--rand_seed', type=int, default=777, help='manual seed')
parser.add_argument('--net_file', type=str,
choices=['wave_net'],
required=True,
help='net file')
parser.add_argument('--start_iter', type=int, default=0)
args = parser.parse_args()
prepare_seed(args.rand_seed)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
save_dir = Path(str(args.save_dir) + str(args.net_file) + '_' + str(args.max_iter))
save_dir.mkdir(exist_ok=True, parents=True)
print(f'=> save_dir: {str(save_dir)}')
# log_dir = Path(args.log_dir)
# log_dir.mkdir(exist_ok=True, parents=True)
writer = SummaryWriter(log_dir=str(save_dir))
net = import_module(args.net_file)
decoder = net.WaveDecoder()
vgg = net.WaveEncoder()
vgg.load_state_dict(torch.load(args.vgg))
network = net.Net(vgg, decoder)
if args.start_iter > 0:
print("Loading state after {:d} iterations".format(args.start_iter + 0))
states = torch.load(save_dir / 'ckpt_iter_{:d}.pth.tar'.format(args.start_iter))
network.decoder.load_state_dict(states['decoder_state_dict'])
network.safin4.load_state_dict(states['safin4_state_dict'])
network.safin3.load_state_dict(states['safin3_state_dict'])
network.train()
network.to(device)
content_tf = train_transform()
style_tf = train_transform()
assert os.path.exists(args.content_dir), args.content_dir
assert os.path.exists(args.style_dir), args.style_dir
content_dataset = FlatFolderDataset(args.content_dir, content_tf)
style_dataset = FlatFolderDataset(args.style_dir, style_tf)
print(f"length of content_dataset:{len(content_dataset)}")
print(f"length of style_dataset: {len(style_dataset)}")
content_iter = iter(data.DataLoader(
content_dataset, batch_size=args.batch_size,
sampler=InfiniteSamplerWrapper(content_dataset),
num_workers=args.n_threads))
style_iter = iter(data.DataLoader(
style_dataset, batch_size=args.batch_size,
sampler=InfiniteSamplerWrapper(style_dataset),
num_workers=args.n_threads))
if hasattr(network, 'safin4'):
params = list(network.decoder.parameters())+list(network.safin4.parameters())+\
list(network.safin3.parameters())
optimizer = torch.optim.Adam(params, lr=args.lr)
print('=> training safin')
for i in tqdm(range(args.max_iter)):
adjust_learning_rate(optimizer, iteration_count=i)
content_images = next(content_iter).to(device)
style_images = next(style_iter).to(device)
network.train()
loss_c, loss_s = network(content_images, style_images)
loss_c = args.content_weight * loss_c
loss_s = args.style_weight * loss_s
loss = loss_c + loss_s
optimizer.zero_grad()
loss.backward()
optimizer.step()
writer.add_scalar('loss_content', loss_c.item(), i + 1)
writer.add_scalar('loss_style', loss_s.item(), i + 1)
if (i + 1) % args.save_model_interval == 0 or (i + 1) == args.max_iter or i == 0:
if hasattr(network, 'safin4'):
states = {
'decoder_state_dict': network.decoder.state_dict(),
'safin4_state_dict': network.safin4.state_dict(),
'safin3_state_dict': network.safin3.state_dict()
}
else:
states = {
'decoder_state_dict': network.decoder.state_dict(),
}
torch.save(states, save_dir /
'ckpt_iter_{:d}.pth.tar'.format(i + 1))
with torch.no_grad():
network.eval()
if hasattr(network, 'safin4'):
safin_list = [network.safin3, network.safin4]
output = style_transfer(vgg, network.decoder, content_images, style_images, \
1.0, safin_list)
else :
output = style_transfer(vgg, network.decoder, content_images, style_images, \
1.0, None)
styled_img_grid = make_grid(output, nrow=4, normalize=True, scale_each=True)
reference_img_grid = make_grid(style_images, nrow=4, normalize=True, scale_each=True)
content_img_grid = make_grid(content_images, nrow=4, normalize=True, scale_each=True)
writer.add_image('styled_images', styled_img_grid, i)
writer.add_image('reference_images', reference_img_grid, i)
writer.add_image('content_images', content_img_grid, i)
writer.close()