-
Notifications
You must be signed in to change notification settings - Fork 0
/
jAG2VG.py
319 lines (301 loc) · 15.5 KB
/
jAG2VG.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
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
import pickle
from PIL import Image
import json
import numpy as np
import os
import argparse
def format_ag_to_vg(args):
source_dir = args.source_dir # default="ActionGenome/dataset/ag/"
dest_dir = args.dest_dir # default="AGinVGformat/"
if ~os.path.exists(dest_dir):
os.mkdir(dest_dir)
if ~os.path.exists(dest_dir+'VG_100K'):
os.mkdir(dest_dir+'VG_100K')
file1 = open(source_dir+'annotations/object_bbox_and_relationship.pkl', 'rb')
file2 = open(source_dir+'annotations/person_bbox.pkl', 'rb')
obj_rel_file = pickle.load(file1)
person_file = pickle.load(file2)
obj_list = ['person', 'bag', 'bed', 'blanket', 'book', 'box', 'broom', 'chair', 'closet/cabinet', 'clothes',
'cup/glass/bottle', 'dish', 'door', 'doorknob', 'doorway', 'floor', 'food', 'groceries', 'laptop',
'light', 'medicine', 'mirror', 'paper/notebook', 'phone/camera', 'picture', 'pillow', 'refrigerator',
'sandwich', 'shelf', 'shoe', 'sofa/couch', 'table', 'television', 'towel', 'vacuum', 'window']
rel_list = ['looking_at', 'not_looking_at', 'unsure', 'above', 'beneath', 'in_front_of', 'behind', 'on_the_side_of',
'in', 'carrying', 'covered_by', 'drinking_from', 'eating', 'have_it_on_the_back', 'holding', 'leaning_on',
'lying_on', 'not_contacting', 'other_relationship', 'sitting_on', 'standing_on', 'touching', 'twisting',
'wearing', 'wiping', 'writing_on']
# rel types: 1 = attention, 2 = spatial, 3 = contacting relation
# 0 = unsure, notcontacting or otherrelationship
rel_list_type = [1, 1, 0, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 0, 0, 3, 3, 3, 3, 3, 3, 3]
rel_skip_list = ['unsure', 'not_contacting', 'other_relationship']
image_id = 0
# Output JSON objects / dicts
image_data = []
objects = []
attributes = []
relationships = [] # spatial, attention and contacting relationships
scene_graphs = []
lists_object_relation = {
"objects_list": obj_list,
"relationships_list": rel_list,
"relationships_list_types": rel_list_type
}
# for attribute_synsets.json
attribute_synsets = {"nothing": "nothing.n.01"}
for key in obj_rel_file.keys():
obj_rel_dict_list = obj_rel_file[key]
person_dict = person_file[key]
subj_id = obj_list.index('person') # Person = Subject
subj_bbox = person_dict['bbox'] # XYXY box format (top-left corner and bottom-right corner)
if np.shape(subj_bbox)[0]>0:
subj_bbox = subj_bbox.astype(int).tolist()
subj_bbox[0][2] = subj_bbox[0][2] - subj_bbox[0][0] # to XYWH box format
subj_bbox[0][3] = subj_bbox[0][3] - subj_bbox[0][1]
if os.path.exists(source_dir + 'frames/' + key):
I = Image.open(source_dir + 'frames/' + key)
[w, h] = I.size
key_split = key.split('/')
filename = key_split[0] + '_' + key_split[1]
I.save(dest_dir + 'VG_100K/' + filename)
os.remove(source_dir + 'frames/' + key)
image_id = image_id + 1
# For image_data.json
image_data.append({
"image_id": image_id,
"filename": filename,
"url": None,
"width": w,
"height": h,
"coco_id": None,
"flicker_id": None
})
for i in range(len(obj_rel_dict_list)):
if obj_rel_dict_list[i]['bbox'] is not None:
obj_rel_dict = obj_rel_dict_list[i]
obj_name = obj_rel_dict['class']
obj_id = obj_list.index(obj_name)
obj_bbox = np.array(obj_rel_dict['bbox']) # XYWH box format (top-left corner, width, height)
obj_bbox = obj_bbox.astype(int).tolist()
sp_rel = obj_rel_dict['spatial_relationship']
at_rel = obj_rel_dict['attention_relationship']
cont_rel = obj_rel_dict['contacting_relationship']
visible = obj_rel_dict['visible']
# if relation is visible, format data in dicts and to vars for JSON files
if visible and (any(item not in rel_skip_list for item in sp_rel) or any(item not in rel_skip_list for item in at_rel)):
# for objects.json
objs = []
objs.append({
"object_id": obj_id,
"x": obj_bbox[0],
"y": obj_bbox[1],
"w": obj_bbox[2],
"h": obj_bbox[3],
"name": obj_name,
"synsets": [obj_name + ".n.01"]
})
objs.append({
"object_id": subj_id,
"x": subj_bbox[0][0],
"y": subj_bbox[0][1],
"w": subj_bbox[0][2],
"h": subj_bbox[0][3],
"name": "person",
"synsets": ["person.n.01"]
})
objects.append({
"image_id": image_id,
"filename": filename,
"objects": objs
})
# for attributes.json
attrs = []
attrs.append({
"object_id": obj_id,
"x": obj_bbox[0],
"y": obj_bbox[1],
"w": obj_bbox[2],
"h": obj_bbox[3],
"name": obj_name,
"synsets": [obj_name + ".n.01"],
"attributes": "nothing"
})
attrs.append({
"object_id": subj_id,
"x": subj_bbox[0][0],
"y": subj_bbox[0][1],
"w": subj_bbox[0][2],
"h": subj_bbox[0][3],
"name": "person",
"synsets": ["person.n.01"],
"attributes": "nothing"
})
attributes.append({
"image_id": image_id,
"filename": filename,
"attributes": attrs
})
# for relationships.json
rels = []
for rel in sp_rel:
if rel not in rel_skip_list:
rels.append({
"relationship_id": rel_list.index(rel),
"predicate": rel,
"synsets": [rel + ".v.01"],
"subject": {
"object_id": subj_id,
"x": subj_bbox[0][0],
"y": subj_bbox[0][1],
"w": subj_bbox[0][2],
"h": subj_bbox[0][3],
"name": "person",
"synsets": ["person.n.01"]
},
"object": {
"object_id": obj_id,
"x": obj_bbox[0],
"y": obj_bbox[1],
"w": obj_bbox[2],
"h": obj_bbox[3],
"name": obj_name,
"synsets": [obj_name + ".n.01"]
}
})
for rel in at_rel:
if rel not in rel_skip_list:
rels.append({
"relationship_id": rel_list.index(rel),
"predicate": rel,
"synsets": [rel + ".v.01"],
"subject": {
"object_id": subj_id,
"x": subj_bbox[0][0],
"y": subj_bbox[0][1],
"w": subj_bbox[0][2],
"h": subj_bbox[0][3],
"name": "person",
"synsets": ["person.n.01"]
},
"object": {
"object_id": obj_id,
"x": obj_bbox[0],
"y": obj_bbox[1],
"w": obj_bbox[2],
"h": obj_bbox[3],
"name": obj_name,
"synsets": [obj_name + ".n.01"]
}
})
for rel in cont_rel:
if rel not in rel_skip_list:
rels.append({
"relationship_id": rel_list.index(rel),
"predicate": rel,
"synsets": [rel + ".v.01"],
"subject": {
"object_id": subj_id,
"x": subj_bbox[0][0],
"y": subj_bbox[0][1],
"w": subj_bbox[0][2],
"h": subj_bbox[0][3],
"name": "person",
"synsets": ["person.n.01"]
},
"object": {
"object_id": obj_id,
"x": obj_bbox[0],
"y": obj_bbox[1],
"w": obj_bbox[2],
"h": obj_bbox[3],
"name": obj_name,
"synsets": [obj_name + ".n.01"]
}
})
relationships.append({
"image_id": image_id,
"filename": filename,
"relationships": rels
})
# for scene_graphs.json
sg_rels = []
for rel in sp_rel:
if rel not in rel_skip_list:
sg_rels.append({
"relationship_id": rel_list.index(rel),
"predicate": rel,
"synsets": [rel + ".v.01"],
"subject_id": subj_id,
"object_id": obj_id
})
for rel in at_rel:
if rel not in rel_skip_list:
sg_rels.append({
"relationship_id": rel_list.index(rel),
"predicate": rel,
"synsets": [rel + ".v.01"],
"subject_id": subj_id,
"object_id": obj_id
})
scene_graphs.append({
"image_id": image_id,
"filename": filename,
"objects": objs,
"relationships": sg_rels
})
print(key + ' processed... ' + str(image_id))
else:
print(key + ' not found')
# save data formatted for JSONs to JSON files
with open(dest_dir+'image_data.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(image_data))
else:
print('image_data.json already exists, not updated.')
with open(dest_dir+'objects.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(objects))
else:
print('objects.json already exists, not updated.')
with open(dest_dir+'attributes.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(attributes))
else:
print('attributes.json already exists, not updated.')
with open(dest_dir+'relationships.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(relationships))
else:
print('relationships.json already exists, not updated.')
with open(dest_dir+'scene_graphs.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(scene_graphs))
else:
print('scene_graphs.json already exists, not updated.')
with open(dest_dir+'lists_object_relation.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(lists_object_relation))
else:
print('lists_object_relation.json already exists, not updated.')
with open(dest_dir+'attribute_synsets.json', 'w+') as f:
if len(f.read()) == 0:
f.write(json.dumps(attribute_synsets))
else:
print('attribute_synsets.json already exists, not updated.')
file1.close()
file2.close()
# Code to write to JSON
# with open('logs.json', 'r+') as f:
# if len(f.read()) == 0:
# f.write(json.dumps(dictionary))
# else:
# f.write(',\n' + json.dumps(dictionary))
# Code to read from JSON
# with open('logs.json') as f:
# g = json.load(f)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Format AG dataset to VG format")
parser.add_argument("--source_dir", default="ActionGenome/dataset/ag/",
help="Folder containing AG videos, frames and annotations folders.")
parser.add_argument("--dest_dir", default="AGinVGformat/",
help="Folder where VG images and JSON files will be saved.")
args = parser.parse_args()
format_ag_to_vg(args)