-
Notifications
You must be signed in to change notification settings - Fork 0
/
text2img_model.py
290 lines (252 loc) · 9.32 KB
/
text2img_model.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
import torch
import torch.nn as nn
from DAMSM import ImageEncoder, TextEncoder, BertEncoder
from modules.models import (
Discriminator64,
Discriminator128,
Discriminator256,
Generator
)
from utils import init_weight, freeze_model
class Text2ImgModel(nn.Module):
def __init__(
self,
embedding_dim,
n_tokens,
text_encoder_embd_size,
pretrained_text_encoder_path,
pretrained_image_encoder_path,
pretrained_generator_path,
branch_num,
num_generator_filters,
num_discriminator_filters,
z_dim,
condition_dim,
is_bert_encoder,
use_sagan,
base_size,
device
):
super(Text2ImgModel, self).__init__()
self.z_dim = z_dim
self.is_bert_encoder = is_bert_encoder
self.image_encoder = ImageEncoder(
multimodal_feat_size=embedding_dim
).to(device)
if pretrained_image_encoder_path != '':
state_dict = torch.load(
pretrained_image_encoder_path,
map_location=lambda storage, loc: storage
)
self.image_encoder.load_state_dict(state_dict)
print('Load image encoder from:', pretrained_image_encoder_path)
else:
print('Warning: no pretrained image encoder')
for p in self.image_encoder.parameters():
p.requires_grad = False
self.image_encoder.eval()
if self.is_bert_encoder:
self.text_encoder = BertEncoder(emb_size=embedding_dim).to(device)
else:
self.text_encoder = TextEncoder(
n_tokens=n_tokens, text_feat_size=embedding_dim,
emb_size=text_encoder_embd_size
).to(device)
if pretrained_text_encoder_path != '':
state_dict = \
torch.load(pretrained_text_encoder_path,
map_location=lambda storage, loc: storage)
self.text_encoder.load_state_dict(state_dict)
print('Load text encoder from:', pretrained_text_encoder_path)
else:
print('Warning: no pretrained text encoder')
for p in self.text_encoder.parameters():
p.requires_grad = False
self.text_encoder.eval()
self.generator = Generator(
ngf=num_generator_filters,
emb_dim=embedding_dim,
ncf=condition_dim,
branch_num=branch_num,
device=device,
z_dim=z_dim,
is_sagan=use_sagan,
base_size=base_size,
).to(device)
self.discriminators = []
if branch_num > 0:
self.discriminators.append(Discriminator64(
dim=num_discriminator_filters,
embd_dim=embedding_dim,
is_spectral=use_sagan,
base_size=base_size,
).to(device))
if branch_num > 1:
self.discriminators.append(Discriminator128(
ndf=num_discriminator_filters,
embd_dim=embedding_dim,
is_spectral=use_sagan,
base_size=base_size,
).to(device))
if branch_num > 2:
self.discriminators.append(Discriminator256(
ndf=num_discriminator_filters,
embd_dim=embedding_dim,
is_spectral=use_sagan,
base_size=base_size,
).to(device))
self.generator.apply(init_weight)
for i in range(len(self.discriminators)):
self.discriminators[i].apply(init_weight)
if pretrained_generator_path != '':
state_dict = torch.load(
pretrained_generator_path,
map_location=lambda storage, loc: storage
)
self.generator.load_state_dict(state_dict)
print('Load generator from: ', pretrained_generator_path)
def forward(self, captions, cap_lens, noise, masks):
if not self.is_bert_encoder:
words_embeddings, sentence_embedding = \
self.text_encoder(captions, cap_lens)
else:
words_embeddings, sentence_embedding = \
self.text_encoder(captions, cap_lens, masks)
mask = (captions == 0)
num_words = words_embeddings.size(2)
if mask.size(1) > num_words:
mask = mask[:, :num_words]
fake_images, attention_maps, mu, logvar = self.generator(
noise,
sentence_embedding,
words_embeddings,
mask
)
return fake_images, mu, logvar, sentence_embedding, words_embeddings
def save_model_ckpt(self, epoch, path):
torch.save({
'epoch': epoch,
'img_enc': self.image_encoder.state_dict(),
'txt_enc': self.text_encoder.state_dict(),
'discriminator': [d.state_dict() for d in self.discriminators],
'generator': self.generator.state_dict()
}, path)
def load_model_ckpt(self, path):
# Load
weights = torch.load(path)
self.image_encoder.load_state_dict(weights['img_enc'])
self.text_encoder.load_state_dict(weights['txt_enc'])
# Freeze parameters
freeze_model(self.image_encoder)
freeze_model(self.text_encoder)
self.image_encoder.eval()
self.text_encoder.eval()
for i, d in enumerate(self.discriminators):
d.load_state_dict(weights['discriminator'][i])
self.generator.load_state_dict(weights['generator'])
return weights['epoch']
class Text2ImgEvalModel(nn.Module):
def __init__(
self,
embedding_dim,
n_tokens,
text_encoder_embd_size,
branch_num,
num_generator_filters,
z_dim,
condition_dim,
is_bert_encoder,
use_sagan,
base_size,
device,
pretrained_ckpt=None,
pretrained_text_encoder_path=None,
pretrained_generator_path=None
):
super(Text2ImgEvalModel, self).__init__()
self.z_dim = z_dim
self.is_bert_encoder = is_bert_encoder
if self.is_bert_encoder:
self.text_encoder = BertEncoder(emb_size=embedding_dim).to(device)
else:
self.text_encoder = TextEncoder(
n_tokens=n_tokens, text_feat_size=embedding_dim,
emb_size=text_encoder_embd_size
).to(device)
self.generator = Generator(
ngf=num_generator_filters,
emb_dim=embedding_dim,
ncf=condition_dim,
branch_num=branch_num,
device=device,
z_dim=z_dim,
is_sagan=use_sagan,
base_size=base_size,
).to(device)
self.load_model(
pretrained_ckpt,
pretrained_text_encoder_path,
pretrained_generator_path
)
def forward(self, captions, cap_lens, noise, masks):
if not self.is_bert_encoder:
words_embeddings, sentence_embedding = \
self.text_encoder(captions, cap_lens)
else:
words_embeddings, sentence_embedding = \
self.text_encoder(captions, cap_lens, masks)
mask = (captions == 0)
num_words = words_embeddings.size(2)
if mask.size(1) > num_words:
mask = mask[:, :num_words]
fake_images, attention_maps, mu, logvar = self.generator(
noise,
sentence_embedding,
words_embeddings,
mask
)
return fake_images, mu, logvar, sentence_embedding, words_embeddings
def load_model(self, ckpt, text_enc, gen):
if ckpt is None and text_enc is None and gen is None:
raise FileNotFoundError("Set path to load the model")
if ckpt is not None and (text_enc is not None or gen is not None):
raise ValueError(
"Specify just one way for loading:"
"checkpoint path or two separate files"
"for text encoder and generator"
)
if ckpt is not None:
print('Loading from checkpoint')
weights = torch.load(ckpt)
self.text_encoder.load_state_dict(weights['txt_enc'])
self.generator.load_state_dict(weights['generator'])
elif gen is not None and text_enc is not None:
print('Loading from separate files')
self.text_encoder.load_state_dict(torch.load(text_enc))
self.generator.load_state_dict(torch.load(gen))
elif gen is None or text_enc is None:
raise FileNotFoundError(
"Specify both generator and text encoder files"
)
self.eval()
freeze_model(self)
if __name__ == '__main__':
DEV = torch.device('cpu')
model = Text2ImgModel(
embedding_dim=256,
n_tokens=20,
text_encoder_embd_size=128,
pretrained_text_encoder_path='',
pretrained_image_encoder_path='',
pretrained_generator_path='',
branch_num=3,
num_generator_filters=32,
num_discriminator_filters=64,
z_dim=100,
condition_dim=128,
is_bert_encoder=False,
base_size=32,
device=DEV,
use_sagan=True
)
print(model)