-
Notifications
You must be signed in to change notification settings - Fork 32
/
train_fast.py
291 lines (234 loc) · 10.5 KB
/
train_fast.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
201
202
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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
import argparse
from pathlib import Path
import torch
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.utils.data as data
from PIL import Image, ImageFile
from torchvision import transforms
from tqdm import tqdm
from template import imagenet_templates
import fast_stylenet
from sampler import InfiniteSamplerWrapper
import clip
from template import imagenet_templates
import torch.nn.functional as F
from torchvision.utils import save_image
from torchvision.transforms.functional import adjust_contrast
cudnn.benchmark = True
Image.MAX_IMAGE_PIXELS = None
ImageFile.LOAD_TRUNCATED_IMAGES = True
def train_transform(crop_size=224):
transform_list = [
transforms.RandomCrop(crop_size),
transforms.ToTensor()
]
return transforms.Compose(transform_list)
def test_transform():
transform_list = [
transforms.Resize(size=(512, 512)),
transforms.ToTensor()
]
return transforms.Compose(transform_list)
def hr_transform():
transform_list = [
transforms.ToTensor()
]
return transforms.Compose(transform_list)
def img_normalize(image):
mean=torch.tensor([0.485, 0.456, 0.406]).to(device)
std=torch.tensor([0.229, 0.224, 0.225]).to(device)
mean = mean.view(1,-1,1,1)
std = std.view(1,-1,1,1)
image = (image-mean)/std
return image
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 load_image(img_path, img_size=None):
image = Image.open(img_path)
if img_size is not None:
image = image.resize((img_size, img_size))
transform = transforms.Compose([
transforms.ToTensor(),
])
image = transform(image)[:3, :, :].unsqueeze(0)
return image
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
def clip_normalize(image):
image = F.interpolate(image,size=224,mode='bicubic')
mean=torch.tensor([0.48145466, 0.4578275, 0.40821073]).to(device)
std=torch.tensor([0.26862954, 0.26130258, 0.27577711]).to(device)
mean = mean.view(1,-1,1,1)
std = std.view(1,-1,1,1)
image = (image-mean)/std
return image
def reverse_normalize(image):
mean=torch.tensor([-0.485/0.229, -0.456/0.224, -0.406/0.225]).to(device)
std=torch.tensor([1./0.229, 1./0.224, 1./0.225]).to(device)
mean = mean.view(1,-1,1,1)
std = std.view(1,-1,1,1)
image = (image-mean)/std
return image
def get_image_prior_losses(inputs_jit):
# COMPUTE total variation regularization loss
diff1 = inputs_jit[:, :, :, :-1] - inputs_jit[:, :, :, 1:]
diff2 = inputs_jit[:, :, :-1, :] - inputs_jit[:, :, 1:, :]
diff3 = inputs_jit[:, :, 1:, :-1] - inputs_jit[:, :, :-1, 1:]
diff4 = inputs_jit[:, :, :-1, :-1] - inputs_jit[:, :, 1:, 1:]
loss_var_l2 = torch.norm(diff1) + torch.norm(diff2) + torch.norm(diff3) + torch.norm(diff4)
return loss_var_l2
parser = argparse.ArgumentParser()
# Basic options
parser.add_argument('--content_dir', type=str, default ='/mnt/F/DIV2K_RW/DIV2K_Realistic_Wild/Train/HR')
parser.add_argument('--test_dir', type=str, default ='./test_set')
parser.add_argument('--hr_dir', type=str)
parser.add_argument('--img_dir', type=str, default ='./test_set')
parser.add_argument('--vgg', type=str, default='models/vgg_normalised.pth')
# training options
parser.add_argument('--save_dir', default='./model_fast',
help='Directory to save the model')
parser.add_argument('--text', default='Fire',
help='text condition')
parser.add_argument('--name', default='none',
help='name')
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=200)
parser.add_argument('--batch_size', type=int, default=4)
parser.add_argument('--content_weight', type=float, default=1.0)
parser.add_argument('--clip_weight', type=float, default=10.0)
parser.add_argument('--tv_weight', type=float, default=1e-4)
parser.add_argument('--glob_weight', type=float, default=1.0)
parser.add_argument('--n_threads', type=int, default=16)
parser.add_argument('--num_test', type=int, default=16)
parser.add_argument('--save_model_interval', type=int, default=200)
parser.add_argument('--save_img_interval', type=int, default=100)
parser.add_argument('--crop_size', type=int, default=224)
parser.add_argument('--thresh', type=float, default=0.7)
parser.add_argument('--decoder', type=str, default='./models/decoder.pth')
args = parser.parse_args()
device = torch.device('cuda')
save_dir = Path(args.save_dir)
save_dir.mkdir(exist_ok=True, parents=True)
decoder = fast_stylenet.decoder
vgg = fast_stylenet.vgg
vgg.load_state_dict(torch.load(args.vgg))
vgg = nn.Sequential(*list(vgg.children())[:31])
decoder.load_state_dict(torch.load(args.decoder))
network = fast_stylenet.Net(vgg, decoder)
network.train()
network.to(device)
clip_model, preprocess = clip.load('ViT-B/32', device, jit=False)
def compose_text_with_templates(text: str, templates=imagenet_templates) -> list:
return [template.format(text) for template in templates]
source = "a Photo"
with torch.no_grad():
prompt = args.text
template_text = compose_text_with_templates(prompt, imagenet_templates)
tokens = clip.tokenize(template_text).to(device)
text_features = clip_model.encode_text(tokens).detach()
text_features = text_features.mean(axis=0, keepdim=True)
text_features /= text_features.norm(dim=-1, keepdim=True)
template_source = compose_text_with_templates(source, imagenet_templates)
tokens_source = clip.tokenize(template_source).to(device)
text_source = clip_model.encode_text(tokens_source).detach()
text_source = text_source.mean(axis=0, keepdim=True)
text_source /= text_source.norm(dim=-1, keepdim=True)
content_tf = train_transform(args.crop_size)
hr_tf = hr_transform()
test_tf = test_transform()
content_dataset = FlatFolderDataset(args.content_dir, content_tf)
test_dataset = FlatFolderDataset(args.test_dir, test_tf)
augment_trans = transforms.Compose([
transforms.RandomPerspective(fill=0, p=1,distortion_scale=0.5),
transforms.Resize(224)
])
content_iter = iter(data.DataLoader(
content_dataset, batch_size=args.batch_size,
sampler=InfiniteSamplerWrapper(content_dataset),
num_workers=args.n_threads))
test_iter = iter(data.DataLoader(
test_dataset, batch_size=args.num_test,
num_workers=args.n_threads))
optimizer = torch.optim.Adam(network.decoder.parameters(), lr=args.lr)
test_images1 = next(test_iter)
test_images1 = test_images1.cuda()
if args.hr_dir is not None:
hr_dataset = FlatFolderDataset(args.hr_dir, hr_tf)
hr_iter = iter(data.DataLoader(
hr_dataset, batch_size=1,
num_workers=args.n_threads))
hr_images = next(hr_iter)
hr_images = hr_images.cuda()
for i in tqdm(range(args.max_iter)):
adjust_learning_rate(optimizer, iteration_count=i)
content_images = next(content_iter).to(device)
loss_c, out_img = network(content_images)
loss_patch = 0
aug_img = []
for it in range(16):
out_aug = augment_trans(out_img)
aug_img.append(out_aug)
aug_img = torch.cat(aug_img,dim=0)
source_features = clip_model.encode_image(clip_normalize(content_images))
source_features /= (source_features.clone().norm(dim=-1, keepdim=True))
image_features = clip_model.encode_image(clip_normalize(aug_img))
image_features /= (image_features.clone().norm(dim=-1, keepdim=True))
img_direction = (image_features-source_features.repeat(16,1))
img_direction /= img_direction.clone().norm(dim=-1, keepdim=True)
text_direction = (text_features-text_source)
text_direction /= text_direction.norm(dim=-1, keepdim=True)
loss_temp = (1- torch.cosine_similarity(img_direction, text_direction.repeat(image_features.size(0),1), dim=1))#.mean()
loss_temp[loss_temp<args.thresh] =0
loss_patch+=loss_temp.mean()
glob_features = clip_model.encode_image(clip_normalize(out_img))
glob_features /= (glob_features.clone().norm(dim=-1, keepdim=True))
glob_direction = (glob_features-source_features)
glob_direction /= glob_direction.clone().norm(dim=-1, keepdim=True)
loss_glob = (1- torch.cosine_similarity(glob_direction, text_direction.repeat(glob_features.size(0),1), dim=1)).mean()
loss_c = args.content_weight * loss_c
reg_tv = args.tv_weight*get_image_prior_losses(out_img)
loss = loss_c + args.clip_weight*loss_patch + args.glob_weight*loss_glob + reg_tv
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i+1)%10==0:
print('loss_content:' + str(loss_c.item()))
print('loss_patch:' + str(loss_patch.item()))
print('loss_dir:' + str(loss_glob.item()))
print('loss_tv:' + str(reg_tv.item()))
if (i + 1) % args.save_model_interval == 0 or (i + 1) == args.max_iter:
state_dict = fast_stylenet.decoder.state_dict()
for key in state_dict.keys():
state_dict[key] = state_dict[key].to(torch.device('cpu'))
torch.save(state_dict, save_dir /
'clip_decoder_iter_{:d}.pth.tar'.format(i + 1))
if (i + 1) % args.save_img_interval ==0 :
with torch.no_grad():
_, test_out1 = network( test_images1)
test_out1 = adjust_contrast(test_out1,1.5)
output_test = torch.cat([test_images1,test_out1],dim=0)
output_name = './output_fast/test1_'+ args.text +'_'+ str(i+1)+'.png'
save_image(output_test, str(output_name),nrow=test_out1.size(0),normalize=True,scale_each=True)
if args.hr_dir is not None:
_, test_out = network(hr_images)
test_out = adjust_contrast(test_out,1.5)
output_name = './output_fast/hr_'+ args.name+'_'+ args.text +'_'+ str(i+1)+'.png'
save_image(test_out, str(output_name),nrow=test_out.size(0),normalize=True,scale_each=True)