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
import json
import os
import random
import time
import dlib
import faiss
import numpy as np
from gm_rpcd.all import bind
from utils.cache import redis_client3, redis_client_db
from utils.es import es_query, es_scan
from utils.images import face_to_vec, url_to_ndarray
base_dir = os.getcwd()
print("base_dir: " + base_dir)
model_dir = os.path.join(base_dir, "_models")
facerec_model_path = os.path.join(model_dir, "dlib_face_recognition_resnet_model_v1.dat")
shape_model_path = os.path.join(model_dir, "shape_predictor_68_face_landmarks.dat")
faiss_index_path = os.path.join(base_dir, "_index", "diary_cover.index")
face_rec = dlib.face_recognition_model_v1(facerec_model_path)
face_detector = dlib.get_frontal_face_detector()
shape_predictor = dlib.shape_predictor(shape_model_path)
FACE_TO_VEC_FUN = lambda img: face_to_vec(img, face_rec, face_detector, shape_predictor)
FAISS_DIARY_INDEX = faiss.read_index(faiss_index_path)
DIARY_AFTER_COVER_FEATURE_KEY = "strategy_embedding:diary:cover:after"
@bind("strategy_embedding/face_similarity/hello")
def hello():
return ["hello", "world"]
# curl "http://172.16.31.17:9200/gm-dbmw-diary/_search?pretty&size=0" -d '
def save_diary_image_info():
q = {
"query": {
"bool": {
"filter": [{
"term": {
"is_online": True
}
}, {
"term": {
"has_cover": True
}
}, {
"term": {
"is_sink": False
}
}, {
"term": {
"has_before_cover": True
}
}, {
"term": {
"has_after_cover": True
}
}, {
"terms": {
"content_level": [6, 5, 4, 3.5, 3]
}
}, {
"term": {
"content_simi_bol_show": 0
}
}, {
"exists": {
"field": "before_cover_url"
}
}]
}
},
"_source": {
"include": ["id", "before_cover_url", "after_cover_url"]
}
}
count = 0
# before_res_dict = {}
after_res_dict = {}
results = es_scan("diary", q)
for item in results:
diary_id = item["_id"]
# before_cover_url = item["_source"]["before_cover_url"] + "-w"
# before_img = url_to_ndarray(before_cover_url)
after_cover_url = item["_source"]["after_cover_url"] + "-w"
img = url_to_ndarray(after_cover_url)
if img.any():
count += 1
print("count: " + str(count) + " " + str(diary_id))
faces = FACE_TO_VEC_FUN(img)
for face in faces:
after_res_dict[diary_id] = face["feature"]
redis_client_db.hmset(DIARY_AFTER_COVER_FEATURE_KEY, after_res_dict)
def save_faiss_index(save_path):
data = redis_client_db.hgetall(DIARY_AFTER_COVER_FEATURE_KEY)
ids = []
features = []
for (k, v) in data.items():
ids.append(str(k, "utf-8"))
features.append(np.array(json.loads(v)))
print("ids: " + str(len(ids)))
ids_np = np.array(ids).astype("int")
features_np = np.array(features).astype("float32")
index = faiss.IndexHNSWFlat(128, 32)
print("trained: " + str(index.is_trained))
index2 = faiss.IndexIDMap(index)
index2.add_with_ids(features_np, ids_np)
faiss.write_index(index2, save_path)
print("faiss index saved")
@bind("strategy_embedding/face_similarity/diary_url")
def get_similar_diary_ids_by_url(url, limit=0.1):
img = url_to_ndarray(url)
if img.any():
faces = FACE_TO_VEC_FUN(img)
for face in faces:
face_feature = np.array(json.loads(face["feature"])).astype("float32")
_scores, _ids = FAISS_DIARY_INDEX.search(np.array([face_feature]), 10)
ids = _ids.flat
scores = _scores.flat
tmp = list(set(zip(ids, scores)))
res = []
for (id, score) in tmp:
if score >= limit:
res.append((id, score))
res.sort(key=lambda x: x[1], reverse=True)
return res
else:
return []
@bind("strategy_embedding/face_similarity/diary_feature")
def get_similar_diary_ids_by_face_features(feature, limit=0.1):
feature = np.array(feature).astype("float32")
_sources, _ids = FAISS_DIARY_INDEX.search(np.array([feature]), 10)
ids = _ids.flat
scores = _sources.flat
tmp = list(set(zip(ids, scores)))
res = []
for (id, score) in tmp:
if score >= limit:
res.append((str(id), float(score)))
res.sort(key=lambda x: x[1], reverse=True)
return res
# def save_diary_similarity(load_file, index_path):
# res_dict = {}
# with open(load_file) as f:
# lines = f.readlines()
# print("lines: " + str(len(lines)))
# count = 0
# for line in lines:
# count += 1
# tmp = line.split("\t")
# id = tmp[0]
# feature = np.array(json.loads(tmp[1]))
# print("{} {}".format(count, id))
# tup_res = get_similar_diary_ids_by_face_features(feature, index_path, FACE_TO_VEC_FUN)
# if tup_res:
# res_dict[id] = json.dumps(tup_res)
# print("done: " + str(len(res_dict)))
# key = random.choice(list(res_dict.keys()))
# print(key + str(res_dict[key]))
# redis_key = "doris:diary:face_similary"
# redis_client3.hmset(redis_key, res_dict)
if __name__ == "__main__":
begin_time = time.time()
save_diary_image_info()
# save_faiss_index(faiss_index_path)
# imgs = [
# "https://pic.igengmei.com/2020/07/03/1437/1b9975bb0b81-w", "https://pic.igengmei.com/2020/07/01/1812/ca64827a83da-w",
# "https://pic.igengmei.com/2020/07/04/1711/24f4131a9b1e-w", "https://pic.igengmei.com/2020/07/04/1507/e17a995be219-w"
# ]
# for img_url in imgs:
# res = get_similar_diary_ids_by_url(img_url, limit=0.18232107)
# print(res)
# print("@@@@@@@@")
# a = [
# -0.08361373096704483, 0.06760436296463013, 0.10752949863672256, -0.020746365189552307, -0.07035162299871445,
# -0.014547230675816536, -0.043201886117458344, -0.12196271121501923, 0.13929598033428192, -0.1360183209180832,
# 0.23247791826725006, -0.08867999166250229, -0.24177594482898712, -0.05600903555750847, -0.05371646583080292,
# 0.22015368938446045, -0.12883149087429047, -0.0822330191731453, -0.0413128100335598, 0.08704500645399094,
# 0.10081718862056732, -0.03764188289642334, 0.036720920354127884, 0.04766431450843811, -0.0685625970363617,
# -0.38336044549942017, -0.10978807508945465, -0.07328074425458908, -0.023904308676719666, -0.007438751868903637,
# -0.09545779973268509, 0.027364756911993027, -0.1537190079689026, -0.04008519649505615, -0.03581209108233452,
# 0.04322449117898941, -0.05686069279909134, -0.11610691249370575, 0.1640746295452118, -0.004643512889742851,
# -0.34821364283561707, 0.03711444139480591, -0.0026186704635620117, 0.1917344480752945, 0.14298999309539795,
# 0.04084448516368866, 0.06119539216160774, -0.12611950933933258, 0.10941470414400101, -0.20786598324775696,
# 0.03435457497835159, 0.11412393301725388, 0.0602775476872921, 0.054409340023994446, -0.002967053558677435,
# -0.12524624168872833, 0.026284342631697655, 0.08236880600452423, -0.10654348134994507, 0.00403654295951128,
# 0.10716681182384491, -0.08270247280597687, 0.018992319703102112, -0.11595900356769562, 0.18344789743423462,
# 0.0895184576511383, -0.1307670772075653, -0.15750591456890106, 0.11103398352861404, -0.13521818816661835,
# -0.03199139982461929, 0.11129119992256165, -0.17407448589801788, -0.20658859610557556, -0.3114454746246338,
# 0.01914297416806221, 0.39955294132232666, 0.12365783005952835, -0.14545315504074097, -0.03254598751664162,
# -0.10342024266719818, 0.03375910595059395, 0.11272192746400833, 0.21788232028484344, 0.08588762581348419,
# 0.012640122324228287, -0.07646650820970535, -0.043292030692100525, 0.21306097507476807, -0.12407292425632477,
# -0.025112995877861977, 0.2634827196598053, 0.005047444254159927, 0.06562616676092148, -0.07397496700286865,
# 0.06206338107585907, -0.0634055882692337, 0.05882266163825989, -0.05909111723303795, 0.027562778443098068,
# 0.043835900723934174, 0.00407575536519289, -0.007656056433916092, 0.1048622876405716, -0.17822585999965668,
# 0.1303984671831131, -0.021631652489304543, 0.0836174339056015, 0.11956407874822617, 0.007379574701189995,
# -0.07777556777000427, -0.08474794030189514, 0.09585978090763092, -0.21120299398899078, 0.1435444951057434,
# 0.19884724915027618, 0.07154559344053268, 0.06259742379188538, 0.10118959099054337, 0.10188969224691391,
# -0.015351934358477592, -0.04335442930459976, -0.26258283853530884, -0.021509556099772453, 0.12185295671224594,
# -0.011788002215325832, 0.01337978895753622, -0.008025042712688446
# ]
# res = get_similar_diary_ids_by_face_features(a)
# print(res)
print("total cost: {:.2f}mins".format((time.time() - begin_time) / 60))