-
Notifications
You must be signed in to change notification settings - Fork 2
/
tools.py
252 lines (208 loc) · 9.19 KB
/
tools.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
import torch
import torch.nn as nn
import torchvision.transforms as transforms
from torchvision.utils import save_image
import os
import zipfile
from PIL import Image
from tqdm import tqdm
from datetime import datetime
####################################### Train Tools #######################################
# save the checkpoint
def save_checkpoint(encoder, transModule, decoder, optimizer, scheduler, epoch,
log_c, log_s, log_id1, log_id2, log_all, loss_count_interval, save_path):
checkpoint = {
'encoder': encoder.state_dict() if not encoder is None else None,
'transModule': transModule.state_dict() if not transModule is None else None,
'decoder': decoder.state_dict() if not decoder is None else None,
'optimizer': optimizer.state_dict() if not optimizer is None else None,
'scheduler': scheduler.state_dict() if not scheduler is None else None,
'epoch': epoch if not epoch is None else None,
'log_c': log_c if not log_c is None else None,
'log_s': log_s if not log_s is None else None,
'log_id1': log_id1 if not log_id1 is None else None,
'log_id2': log_id2 if not log_id2 is None else None,
'log_all': log_all if not log_all is None else None,
'loss_count_interval': loss_count_interval if not loss_count_interval is None else None
}
torch.save(checkpoint, save_path)
######################################## Test Tools #######################################
def showTorchImage(image):
if len(image.shape) == 4:
image = image.squeeze(0)
mode = transforms.ToPILImage()(image)
plt.imshow(mode)
plt.show()
plt.close()
def zip_dir(zipFile_name, dir_path):
z = zipfile.ZipFile(zipFile_name, 'w', zipfile.ZIP_DEFLATED)
for dirpath, dirnames, filenames in os.walk(dir_path):
for filename in filenames:
z.write(os.path.join(dirpath, filename))
z.close()
def open_img_to_pt(img_path, transform=transforms.ToTensor()):
img = Image.open(img_path)
img_pt = transform(img).unsqueeze(dim=0)
return img_pt
def content_style_transTo_pt(i_c_path, i_s_path, i_c_size=None):
"""Resize the pics of arbitrary size to the shape of content image
"""
i_c_pil = Image.open(i_c_path)
i_s_pil = Image.open(i_s_path)
if not i_c_size is None:
i_c_tf = transforms.Compose([
transforms.Resize(i_c_size),
transforms.ToTensor()
])
else:
i_c_tf = transforms.Compose([
transforms.ToTensor()
])
i_s_size = min(i_c_pil.size[1], i_c_pil.size[0])
i_s_tf = transforms.Compose([
transforms.Resize(i_s_size),
transforms.ToTensor()
])
i_c_pt = i_c_tf(i_c_pil).unsqueeze(dim=0)
i_s_pt = i_s_tf(i_s_pil).unsqueeze(dim=0)
return i_c_pt, i_s_pt
@torch.no_grad()
def save_sample_imgs(network, samples_path, img_saved_path, device=torch.device('cpu')):
"""Test and save samples imgs (Fixed Size)
Args:
network : Model that tested
samples_path : Path where the samples saved
Required two sub-dirs named 'Content' and 'Style'
img_saved_path: Path to save the results
"""
sample_dict = {
'1': [1,2,5,6],
'2': [3,6,9],
'3': [4,6,9],
'4': [1,8,9],
'5': [1,6,8],
'6': [1,6,7],
'7': [1,6,9],
'8': [1,6,8],
'9': [1,6,7],
}
print('Image generation starts:')
for i_c_num in tqdm(sample_dict.keys()):
output_imgs = torch.tensor([])
for i_s_num in sample_dict[i_c_num]:
i_c = open_img_to_pt(os.path.join(samples_path, f'Content/{i_c_num}.png')).to(device)
i_s = open_img_to_pt(os.path.join(samples_path, f'Style/{i_s_num}.png')).to(device)
i_cs = network(i_c, i_s)
output_img = torch.cat((i_c.cpu(), i_s.cpu(), i_cs.cpu()), dim=0)
output_imgs = torch.cat((output_imgs, output_img), dim=0)
output_name = os.path.join(img_saved_path, f'test_{i_c_num}.png')
save_image(output_imgs, output_name, nrow=3)
@torch.no_grad()
def save_sample_imgs_arbitrarySize(network, samples_path, img_saved_path, device=torch.device('cpu')):
"""Test and save samples imgs (Arbitrary Size)
Args:
network : Model that tested
samples_path : Path where the samples saved
Required two sub-dirs named 'Content' and 'Style'
img_saved_path: Path to save the results
"""
sample_dict = {
'1': [1,2,3,4,5,6,7,8,9,10,11],
'2': [1,2,3,4,5,6,7,8,9,10,11],
}
print('Image generation starts:')
for i_c_num in tqdm(sample_dict.keys()):
i_cs = torch.tensor([])
output_imgs = torch.tensor([])
i_c_path = os.path.join(samples_path, f'Content/{i_c_num}.jpg')
for i_s_num in sample_dict[i_c_num]:
i_s_path = os.path.join(samples_path, f'Style/{i_s_num}.jpg')
i_c, i_s = content_style_transTo_pt(i_c_path, i_s_path)
i_cs = network(i_c.to(device), i_s.to(device), arbitrary_input=True)
i_s = transforms.CenterCrop((i_c.shape[2], i_c.shape[3]))(i_s)
i_cs = transforms.CenterCrop((i_c.shape[2], i_c.shape[3]))(i_cs)
output_img = torch.cat((i_c.cpu(), i_s.cpu(), i_cs.cpu()), dim=0)
output_imgs = torch.cat((output_imgs, output_img), dim=0)
output_name = os.path.join(img_saved_path, f'test_{i_c_num}.jpg')
save_image(output_imgs, output_name, nrow=3)
@torch.no_grad()
def save_transferred_imgs(network, samples_path, img_saved_path, device=torch.device('cpu')):
print('Image generation starts:')
i_c_names = os.listdir(os.path.join(samples_path, 'Content'))
i_s_names = os.listdir(os.path.join(samples_path, 'Style'))
for i_c_name in tqdm(i_c_names):
for i_s_name in tqdm(i_s_names):
i_c_path = os.path.join(samples_path, 'Content', i_c_name)
i_s_path = os.path.join(samples_path, 'Style', i_s_name)
i_c, i_s = content_style_transTo_pt(i_c_path, i_s_path)
i_cs = network(i_c.to(device), i_s.to(device), arbitrary_input=True)
stem_c, suffix_c = os.path.splitext(i_c_name)
stem_s, suffix_s = os.path.splitext(i_s_name)
output_name = os.path.join(img_saved_path, f'{stem_c}_+_{stem_s}.{suffix_c}')
save_image(i_cs, output_name)
@torch.no_grad()
def save_content_leak_imgs(network, samples_path, img_saved_path, rounds=20, device=torch.device('cpu')):
print('Image generation starts:')
i_c_names = os.listdir(os.path.join(samples_path, 'Content'))
i_s_names = os.listdir(os.path.join(samples_path, 'Style'))
for i_c_name in tqdm(i_c_names):
for i_s_name in tqdm(i_s_names):
i_c_path = os.path.join(samples_path, 'Content', i_c_name)
i_s_path = os.path.join(samples_path, 'Style', i_s_name)
i_c, i_s = content_style_transTo_pt(i_c_path, i_s_path)
i_c = i_c.to(device)
i_s = i_s.to(device)
i_cs = i_c
for i in range(rounds):
i_cs = network(i_cs, i_s, arbitrary_input=True)
stem_c, suffix_c = os.path.splitext(i_c_name)
stem_s, suffix_s = os.path.splitext(i_s_name)
output_name = os.path.join(img_saved_path, f'{stem_c}_+_{stem_s}.{suffix_c}')
save_image(i_cs, output_name)
@torch.no_grad()
def caculate_avg_generate_time(network, i_c_path, i_s_path, round=1, device=torch.device('cpu')):
i_c = open_img_to_pt(i_c_path)
i_s = open_img_to_pt(i_s_path)
i_c = i_c.to(device)
i_s = i_s.to(device)
time_start = datetime.now()
for i in range(round):
i_cs = network(i_c, i_s, arbitrary_input=True)
time_end = datetime.now()
avg_generate_time = ((time_end-time_start).seconds + (time_end-time_start).microseconds/1000000) / round
return avg_generate_time
@torch.no_grad()
def caculate_avg_generate_time_multiple(network, samples_path, device=torch.device('cpu')):
i_c_names = os.listdir(os.path.join(samples_path, 'Content'))
i_s_names = os.listdir(os.path.join(samples_path, 'Style'))
nums = len(i_c_names) * len(i_s_names)
time_start = datetime.now()
for i_c_name in i_c_names:
for i_s_name in i_s_names:
i_c_path = os.path.join(samples_path, 'Content', i_c_name)
i_s_path = os.path.join(samples_path, 'Style', i_s_name)
i_c, i_s = content_style_transTo_pt(i_c_path, i_s_path)
i_c = i_c.to(device)
i_s = i_s.to(device)
i_cs = network(i_c, i_s, arbitrary_input=True)
time_end = datetime.now()
avg_generate_time = ((time_end-time_start).seconds + (time_end-time_start).microseconds/1000000) / nums
return avg_generate_time
class Sample_Test_Net(nn.Module):
def __init__(self, encoder, decoder, transModule, patch_size=8):
super(Sample_Test_Net, self).__init__()
self.encoder = encoder
self.decoder = decoder
self.transModule = transModule
self.patch_size = patch_size
def forward(self, i_c, i_s, arbitrary_input=False):
_, _, H, W = i_c.size()
self.decoder.img_H = H
self.decoder.img_W = W
f_c = self.encoder(i_c, arbitrary_input)
f_s = self.encoder(i_s, arbitrary_input)
f_c, f_c_reso = f_c[0], f_c[2]
f_s, f_s_reso = f_s[0], f_s[2]
f_cs = self.transModule(f_c, f_s)
i_cs = self.decoder(f_cs, f_c_reso)
return i_cs