-
Notifications
You must be signed in to change notification settings - Fork 0
/
demo_face_recog.py
161 lines (146 loc) · 6.8 KB
/
demo_face_recog.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
# author: zerg
# libs
from multiprocessing import Process, Queue
import numpy as np
import cv2
import process
import copy
import glob
import os
import sys
from retinaface import RetinaFace
import face_model
sys.path.append(os.path.join(os.path.dirname(__file__), 'common'))
import face_preprocess
# params
num_skip = 6 # for speed reason
name_window = 'frame'
# path_video = 'rtsp://192.168.3.34:554/live/ch4'
path_video = 'rtsp://192.168.3.225:554/ch4'
# path_video = '/media/manu/samsung/videos/at2021/mp4/Video1.mp4'
model_face_detect_path = '/home/manu/tmp/mobilenet_v1_0_25/retina'
warmup_img_path = '/media/manu/samsung/pics/material3000_1920x1080.jpg' # image size should be same as actual input
gpuid = 0
thresh = 0.3
scales = [1.0]
flip = False
face_recog_debug_dir = '/home/manu/tmp/demo_snapshot/'
face_dataset_dir = '/media/manu/samsung/pics/人脸底图'
model_face_recog_path = '/home/manu/tmp/t3m0.4/model,28'
face_recog_sim_th = 0.35
# face_recog_dist_th = 2.0
if __name__ == '__main__':
print('face detect init start ...')
detector = RetinaFace(model_face_detect_path, 0, gpuid, 'net3')
img = cv2.imread(warmup_img_path)
print('face detect init done')
print('face recog init start ...')
face_recog_dataset = []
model = face_model.FaceModel(gpuid, model_face_recog_path)
out_dir = face_recog_debug_dir
os.system('rm %s -rvf' % out_dir)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
for path_img in glob.glob(os.path.join(face_dataset_dir, '*.jpg')):
_, img_name = os.path.split(path_img)
img_name = img_name.replace('.jpg', '')
stu_id, stu_name = img_name.split('_')
img = cv2.imread(path_img)
h, w, c = img.shape
max_l = max(h, w)
scales_reg = [1.0]
if max_l > 1920: # can not detect faces on some large input images
scales_reg = [0.5]
faces, landmarks = detector.detect(img, 0.8, scales=scales_reg, do_flip=flip) # using high detect th for reg
assert len(faces) == 1 # TODO
# face align and feature extract
bbox = faces
points = np.squeeze(landmarks).transpose().reshape(1, 10)
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
img_aligned = face_preprocess.preprocess(img, bbox, points, image_size='112,112')
out_path = os.path.join(out_dir, stu_name + '.jpg')
cv2.imwrite(out_path, img_aligned)
img_aligned = cv2.cvtColor(img_aligned, cv2.COLOR_BGR2RGB)
img_aligned = np.transpose(img_aligned, (2, 0, 1))
feat = model.get_feature(img_aligned)
item = (stu_id, stu_name, feat)
face_recog_dataset.append(item)
print('record student %s with id %s' % (stu_name, stu_id))
print('face recog init done')
print('warm up start ...')
# warm up
for _ in range(10):
faces, landmarks = detector.detect(img, thresh, scales=scales, do_flip=flip)
print('warm up done')
q_decoder = Queue()
p_decoder = Process(target=process.process_decoder, args=(q_decoder, path_video, num_skip), daemon=True)
p_decoder.start()
cv2.namedWindow(name_window, cv2.WINDOW_NORMAL)
cv2.resizeWindow(name_window, 960, 540)
face_recog_aligned_save_idx = 0
while True:
item_frame = q_decoder.get()
frame_org = item_frame[0]
frame = copy.deepcopy(frame_org)
h, w, c = frame_org.shape
if w > 1920:
scales = [0.5]
faces, landmarks = detector.detect(frame, thresh, scales=scales, do_flip=flip)
if faces is not None:
print('find', faces.shape[0], 'faces')
# recognition
for i in range(faces.shape[0]):
box = faces[i].astype(int)
bbox = faces[i]
landmarks_recog = landmarks[i]
points = landmarks_recog.transpose().reshape(1, 10)
bbox = bbox[0:4]
points = points[0, :].reshape((2, 5)).T
img_aligned = face_preprocess.preprocess(frame_org, bbox, points, image_size='112,112')
img_aligned_write = img_aligned
img_aligned = cv2.cvtColor(img_aligned, cv2.COLOR_BGR2RGB)
img_aligned = np.transpose(img_aligned, (2, 0, 1))
feat = model.get_feature(img_aligned)
[sim_highest, stu_name_highest, isfind] = [0, 'null', False]
for stu_id, stu_name, feat_ref in face_recog_dataset:
sim = np.dot(feat_ref, feat.T) # sim is wired
if sim > sim_highest:
sim_highest = sim
stu_name_highest = stu_name
# dist = np.sum(np.square(feat_ref - feat))
# if dist < face_recog_dist_th:
if sim > face_recog_sim_th:
# info = stu_name + ' with dist ' + '%f' % dist
info = stu_name + ' ' + '%f' % sim
img = cv2.putText(frame, info, (box[0], box[1]), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 0, 255), 2)
# save aligned image for debug reason
out_dir = face_recog_debug_dir
# out_path = os.path.join(out_dir,
# '%s_%d_%f' % (stu_name, face_recog_aligned_save_idx, dist) + '.jpg')
out_path = os.path.join(out_dir,
'%s_%d_%f' % (stu_name, face_recog_aligned_save_idx, sim) + '.jpg')
cv2.imwrite(out_path, img_aligned_write)
face_recog_aligned_save_idx += 1
isfind = True
if not isfind:
info = stu_name_highest + ' ' + '%f' % sim_highest
img = cv2.putText(frame, info, (box[0], box[1]), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 255, 0), 2)
# plot
for i in range(faces.shape[0]):
# print('score', faces[i][4])
box = faces[i].astype(int)
# color = (255,0,0)
cv2.rectangle(frame, (box[0], box[1]), (box[2], box[3]), (0, 255, 0), 2)
if landmarks is not None:
landmark5 = landmarks[i].astype(int)
# print(landmark.shape)
for l in range(landmark5.shape[0]):
color = (0, 0, 255)
if l == 0 or l == 3:
color = (0, 255, 0)
cv2.circle(frame, (landmark5[l][0], landmark5[l][1]), 1, color, 2)
cv2.imshow(name_window, frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()