File size: 4,835 Bytes
f82a827
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
'''
Rename the Open Images V6 images to adapt pycocotools and produce the corresponding annotations.
The Visual Genome annotations are produced in the same way.
'''
import json
import os

root_path = 'your_path/open-imagev6/' # Download images from Rongjie's repo and unzip it.


with open(root_path +'annotations/'+ 'categories_dict.json') as f:
    categories = json.load(f)
obj_categories = categories['obj']
rel_categories = categories['rel']

categories = []
for idx, i in enumerate(obj_categories):
    category = {'supercategory': i, 'id': idx, 'name': i}
    categories.append(category)

with open(root_path+'annotations/'+'vrd-train-anno.json') as f:
    train_image_list = json.load(f)

counter = 0
new_image_name = 1

images = []
annotations = []
train_rel = {}
#

for row in train_image_list:
    image_path = root_path+'images/'+row['img_fn']+'.jpg' #TODO image_id
    w, h = row['img_size']
    os.rename(image_path, root_path+'images/'+str(new_image_name)+'.jpg')

    image = {'file_name': str(new_image_name)+'.jpg',
              'height': h,
              'width': w,
              'id': new_image_name}
    images.append(image)

    for index, j in enumerate(row['bbox']):

        bbox = [j[0], j[1], j[2]-j[0], j[3]-j[1]] #cxcywh
        area = int(bbox[2] * bbox[3])
        anno_id = counter
        counter = counter + 1

        annotation = {'segmentation': None,
                      'area': area,
                      'bbox': bbox,
                      'iscrowd': 0,
                      'image_id': new_image_name,
                      'id': anno_id,
                      'category_id': row['det_labels'][index]}
        annotations.append(annotation)

    train_rel[new_image_name] = row['rel']

    new_image_name = new_image_name + 1

train_database = {'images': images,
                 'annotations': annotations,
                 'categories': categories}

print('train finish')

with open(root_path+'annotations/'+'vrd-test-anno.json') as f:
    test_image_list = json.load(f)

images = []
annotations = []
test_rel = {}

for row in test_image_list:
    image_path = root_path+'images/'+row['img_fn']+'.jpg' #TODO image_id
    w, h = row['img_size']
    os.rename(image_path, root_path+'images/'+str(new_image_name)+'.jpg')

    image = {'file_name': str(new_image_name)+'.jpg',
              'height': h,
              'width': w,
              'id': new_image_name}
    images.append(image)

    for index, j in enumerate(row['bbox']):

        bbox = [j[0], j[1], j[2]-j[0], j[3]-j[1]] #cxcywh
        area = int(bbox[2] * bbox[3])
        anno_id = counter
        counter = counter + 1

        annotation = {'segmentation': None,
                      'area': area,
                      'bbox': bbox,
                      'iscrowd': 0,
                      'image_id': new_image_name,
                      'id': anno_id,
                      'category_id': row['det_labels'][index]}
        annotations.append(annotation)

    test_rel[new_image_name] = row['rel']

    new_image_name = new_image_name + 1

test_database = {'images': images,
                 'annotations': annotations,
                 'categories': categories}

print('test finish')


with open(root_path+'annotations/'+'vrd-val-anno.json') as f:
    val_image_list = json.load(f)

images = []
annotations = []
val_rel = {}

for row in val_image_list:
    image_path = root_path+'images/'+row['img_fn']+'.jpg' #TODO image_id
    w, h = row['img_size']
    os.rename(image_path, root_path+'images/'+str(new_image_name)+'.jpg')

    image = {'file_name': str(new_image_name)+'.jpg',
              'height': h,
              'width': w,
              'id': new_image_name}
    images.append(image)

    for index, j in enumerate(row['bbox']):

        bbox = [j[0], j[1], j[2]-j[0], j[3]-j[1]] #cxcywh
        area = int(bbox[2] * bbox[3])
        anno_id = counter
        counter = counter + 1

        annotation = {'segmentation': None,
                      'area': area,
                      'bbox': bbox,
                      'iscrowd': 0,
                      'image_id': new_image_name,
                      'id': anno_id,
                      'category_id': row['det_labels'][index]}
        annotations.append(annotation)

    val_rel[new_image_name] = row['rel']

    new_image_name = new_image_name + 1

val_database = {'images': images,
                 'annotations': annotations,
                 'categories': categories}

print('test finish')

rel_database = {'train': train_rel,
                'val': val_rel,
                'test': test_rel,
                'rel_categories': rel_categories}


json.dump(train_database, open('train.json', 'w'))
json.dump(val_database, open('val.json', 'w'))
json.dump(test_database, open('test.json', 'w'))
json.dump(rel_database, open('rel.json', 'w'))