-
Notifications
You must be signed in to change notification settings - Fork 7
Expand file tree
/
Copy pathpre_transforms.py
More file actions
173 lines (128 loc) · 5.65 KB
/
Copy pathpre_transforms.py
File metadata and controls
173 lines (128 loc) · 5.65 KB
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
# coding: utf-8
# In[1]:
import jieba.analyse
jieba.analyse.set_stop_words('static/dataset/stopwords.txt')
from torchvision import transforms
import json, os
import numpy as np
import torch
from PIL import Image, ImageFile
import torchvision.models as models
from torch.autograd import Variable
dataset_dir = 'static/dataset/arch'
max_num = 1000
# In[2]:
image_transform = transforms.Compose([
transforms.Scale([224, 224]),
transforms.ToTensor(),
transforms.Normalize(mean = [ 0.485, 0.456, 0.406 ],
std = [ 0.229, 0.224, 0.225 ]),
])
resnet = models.resnet152(pretrained=True)
resnet.fc = torch.nn.Dropout(p=0)
resnet = resnet.eval()
resnet = resnet.cuda()
def normalize(v):
norm=np.linalg.norm(v)
if norm==0:
return v
return v/norm
def pre_transforms():
ImageFile.LOAD_TRUNCATED_IMAGES = True
print 'Pre-transforming train ...'
json_file = 'annotations/data_pair_train.json'
train_data_pair = json.load(open(os.path.join(dataset_dir, json_file), 'r'))
captions_train, images_train = [], []
for k, item in enumerate(train_data_pair):
if k % 100 == 0:
print 'Processing %d/%d' %(k, len(train_data_pair))
caption = ' '.join(jieba.analyse.extract_tags(item['caption'], topK=20, withWeight=False, allowPOS=()))
if len(caption) == 0:
continue
captions_train.append(caption.encode('utf8')+'\n')
img_vec = image_transform(Image.open(item['img_path']).convert('RGB')).unsqueeze(0)
img_vec = resnet(Variable(img_vec.cuda())).data.squeeze(0).cpu().numpy()
images_train.append(img_vec)
if k > max_num:
break
with open('arch_train_caps.txt', 'w') as f_write:
f_write.writelines(captions_train)
images_train = np.asarray(images_train, dtype=np.float32)
images_train = normalize(images_train)
np.save('arch_train_ims.npy', images_train)
print 'Pre-transforming train Done'
print 'Pre-transforming dev ...'
json_file = 'annotations/data_pair_val.json'
dev_data_pair = json.load(open(os.path.join(dataset_dir, json_file), 'r'))
captions_dev, images_dev = [], []
caps_obj_id, imgs_obj_id = [], []
caps_url, imgs_url, imgs_path = [], [], []
for k, item in enumerate(dev_data_pair):
if k % 100 == 0:
print 'Processing %d/%d' %(k, len(dev_data_pair))
if item['obj_id'] not in caps_obj_id:
caption = ' '.join(jieba.analyse.extract_tags(item['caption'], topK=20, withWeight=False, allowPOS=()))
if len(caption) == 0:
continue
captions_dev.append(caption.encode('utf8')+'\n')
caps_obj_id.append(item['obj_id'])
caps_url.append(item['url'])
img_vec = image_transform(Image.open(item['img_path']).convert('RGB')).unsqueeze(0)
img_vec = resnet(Variable(img_vec.cuda())).data.squeeze(0).cpu().numpy()
images_dev.append(img_vec)
imgs_obj_id.append(item['obj_id'])
imgs_url.append(item['url'])
imgs_path.append(item['img_path'])
if k > max_num:
break
with open('arch_dev_caps.txt', 'w') as f_write:
f_write.writelines(captions_dev)
json.dump(caps_url, open('arch_dev_caps_url.json', 'w'))
json.dump(imgs_url, open('arch_dev_imgs_url.json', 'w'))
json.dump(imgs_path, open('arch_dev_imgs_path.json', 'w'))
images_dev = np.asarray(images_dev, dtype=np.float32)
images_dev = normalize(images_dev)
np.save('arch_dev_ims.npy', images_dev)
caps_obj_id = np.asarray(caps_obj_id, dtype=np.float32)
imgs_obj_id = np.asarray(imgs_obj_id, dtype=np.float32)
np.save('arch_dev_caps_id.npy', caps_obj_id)
np.save('arch_dev_imgs_id.npy', imgs_obj_id)
print 'Pre-transforming dev Done'
print 'Pre-transforming test ...'
test_data_pair = train_data_pair + dev_data_pair
captions_test, images_test = [], []
caps_obj_id_test, imgs_obj_id_test = [], []
caps_url_test, imgs_url_test, imgs_path_test = [], [], []
for k, item in enumerate(test_data_pair):
if k % 100 == 0:
print 'Processing %d/%d' %(k, len(test_data_pair))
if item['obj_id'] not in caps_obj_id_test:
caption = ' '.join(jieba.analyse.extract_tags(item['caption'], topK=20, withWeight=False, allowPOS=()))
if len(caption) == 0:
continue
captions_test.append(caption.encode('utf8')+'\n')
caps_obj_id_test.append(item['obj_id'])
caps_url_test.append(item['url'])
img_vec = image_transform(Image.open(item['img_path']).convert('RGB')).unsqueeze(0)
img_vec = resnet(Variable(img_vec.cuda())).data.squeeze(0).cpu().numpy()
images_test.append(img_vec)
imgs_obj_id_test.append(item['obj_id'])
imgs_url_test.append(item['url'])
imgs_path_test.append(item['img_path'])
if k > max_num:
break
with open('arch_test_caps.txt', 'w') as f_write:
f_write.writelines(captions_test)
json.dump(caps_url_test, open('arch_test_caps_url.json', 'w'))
json.dump(imgs_url_test, open('arch_test_imgs_url.json', 'w'))
json.dump(imgs_path_test, open('arch_test_imgs_path.json', 'w'))
images_test = np.asarray(images_test, dtype=np.float32)
images_test = normalize(images_test)
np.save('arch_test_ims.npy', images_test)
caps_obj_id_test = np.asarray(caps_obj_id_test, dtype=np.float32)
imgs_obj_id_test = np.asarray(imgs_obj_id_test, dtype=np.float32)
np.save('arch_test_caps_id.npy', caps_obj_id_test)
np.save('arch_test_imgs_id.npy', imgs_obj_id_test)
print 'Pre-transforming test Done'
if __name__ == "__main__":
pre_transforms()