face_detecter.py
8.18 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
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
import os
import numpy as np
import MNN
import cv2
import logging
from retinaface import PriorBox
def py_cpu_nms(dets, thresh):
"""Pure Python NMS baseline."""
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1)
order = scores.argsort()[::-1]
keep = []
while order.size > 0:
i = order[0]
keep.append(i)
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
ovr = inter / (areas[i] + areas[order[1:]] - inter)
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
return keep
def decode_landm(pre, priors, variances):
landms = np.concatenate((priors[:, :2] + pre[:, :2] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 2:4] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 4:6] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 6:8] * variances[0] * priors[:, 2:],
priors[:, :2] + pre[:, 8:10] * variances[0] * priors[:, 2:]), 1)
return landms
def decode(loc, priors, variances):
boxes = np.concatenate((
priors[:, :2] + loc[:, :2] * variances[0] * priors[:, 2:],
priors[:, 2:] * np.exp(loc[:, 2:] * variances[1])), 1)
boxes[:, :2] -= boxes[:, 2:] / 2
boxes[:, 2:] += boxes[:, :2]
return boxes
class Face_Detector(object):
def __init__(self, model_path):
logging.info('******** Start Init Face Detector ********')
self.det_interpreter = MNN.Interpreter(model_path)
self.det_session = self.det_interpreter.createSession()
self.det_input_tensor = self.det_interpreter.getSessionInput(self.det_session)
logging.info('******** Success Init Face Detector ********')
def detect(self, frame, thr):
logging.info('******** Start Face Detect ********')
input_size = 320
img = cv2.resize(frame, (input_size, input_size))
img = np.float32(img)
im_height, im_width, _ = img.shape
scale = np.array([img.shape[1], img.shape[0], img.shape[1], img.shape[0]])
scale1 = np.array([img.shape[1], img.shape[0], img.shape[1], img.shape[0],
img.shape[1], img.shape[0], img.shape[1], img.shape[0],
img.shape[1], img.shape[0]])
w_r = input_size/frame.shape[1]
h_r = input_size/frame.shape[0]
confidence_threshold = 0.02
vis_threshold = 0.5
nms_threshold = 0.4
keep_top_k = 100
variance = [0.1, 0.2]
img -= (104, 117, 123)
img = img.transpose(2, 0, 1)
img = np.expand_dims(img, axis=0)
input_tensor = MNN.Tensor((1, 3, input_size, input_size), MNN.Halide_Type_Float, img, MNN.Tensor_DimensionType_Caffe)
self.det_input_tensor.copyFrom(input_tensor)
self.det_interpreter.runSession(self.det_session)
bbox_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output0')
conf_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output1')
landmark_output_tensor = self.det_interpreter.getSessionOutput(self.det_session, 'output2')
bbox_output = bbox_output_tensor.getData()
conf_output = conf_output_tensor.getData()
landmark_output = landmark_output_tensor.getData()
norm_confs = list()
for i in range(int(len(conf_output)/2)):
norm_confs.append([conf_output[i * 2 + 0], conf_output[i * 2 + 1]])
norm_bboxes = list()
for i in range(int(len(conf_output)/2)):
norm_bboxes.append([bbox_output[i * 4 + 0], bbox_output[i * 4 + 1], bbox_output[i * 4 + 2], bbox_output[i * 4 + 3]])
norm_landmarks = list()
for i in range(int(len(conf_output)/2)):
norm_landmarks.append([landmark_output[i * 10 + 0], landmark_output[i * 10 + 1],
landmark_output[i * 10 + 2], landmark_output[i * 10 + 3],
landmark_output[i * 10 + 4], landmark_output[i * 10 + 5],
landmark_output[i * 10 + 6], landmark_output[i * 10 + 7],
landmark_output[i * 10 + 8], landmark_output[i * 10 + 9]])
norm_confs = np.array(norm_confs)
norm_bboxes = np.array(norm_bboxes)
norm_landmarks = np.array(norm_landmarks)
priorbox = PriorBox(image_size=(im_height, im_width))
priors = priorbox.forward()
scores = norm_confs[:, 1]
boxes = decode(norm_bboxes, priors, variance)
boxes = boxes * scale
landms = decode_landm(norm_landmarks, priors, variance)
landms = landms * scale1
# ignore low scores
inds = np.where(scores > confidence_threshold)[0]
boxes = boxes[inds]
landms = landms[inds]
scores = scores[inds]
# keep top-K before NMS
order = scores.argsort()[::-1][:keep_top_k]
boxes = boxes[order]
landms = landms[order]
scores = scores[order]
# do NMS
dets = np.hstack((boxes, scores[:, np.newaxis])).astype(np.float32, copy=False)
keep = py_cpu_nms(dets, nms_threshold)
dets = dets[keep, :]
landms = landms[keep]
# keep top-K faster NMS
dets = dets[:keep_top_k, :]
landms = landms[:keep_top_k, :]
dets = np.concatenate((dets, landms), axis=1)
face_bboxes = []
face_landmarks = []
max_area = float('-inf')
max_index = 0
i = 0
for b in dets:
if b[4] < vis_threshold:
continue
resize_b = []
x1 = int(b[0] / w_r)
y1 = int(b[1] / h_r)
x2 = int(b[2] / w_r)
y2 = int(b[3] / h_r)
x3 = int(b[5] / w_r)
y3 = int(b[6] / h_r)
x4 = int(b[7] / w_r)
y4 = int(b[8] / h_r)
x5 = int(b[9] / w_r)
y5 = int(b[10] / h_r)
x6 = int(b[11] / w_r)
y6 = int(b[12] / h_r)
x7 = int(b[13] / w_r)
y7 = int(b[14] / h_r)
resize_b = [x1, y1, x2, y2, 0, x3, y3, x4, y4, x5, y5, x6, y6, x7, y7]
# cv2.rectangle(frame, (resize_b[0], resize_b[1]), (resize_b[2], resize_b[3]), (0, 0, 255), 2)
# cv2.circle(frame, (resize_b[5], resize_b[6]), 1, (0, 0, 255), 4)
# cv2.circle(frame, (resize_b[7], resize_b[8]), 1, (0, 255, 255), 4)
# cv2.circle(frame, (resize_b[9], resize_b[10]), 1, (255, 0, 255), 4)
# cv2.circle(frame, (resize_b[11], resize_b[12]), 1, (0, 255, 0), 4)
# cv2.circle(frame, (resize_b[13], resize_b[14]), 1, (255, 0, 0), 4)
area = (resize_b[2] - resize_b[0]) * (resize_b[3] - resize_b[1])
if area > max_area:
max_area = area
max_index = i
i += 1
face_bboxes.append([resize_b[0], resize_b[1], resize_b[2], resize_b[3]])
face_landmarks.append([(resize_b[5], resize_b[6]),
(resize_b[7], resize_b[8]),
(resize_b[9], resize_b[10]),
(resize_b[11], resize_b[12]),
(resize_b[13], resize_b[14])])
# import time
# cv2.imwrite('results/0.jpg', frame)
return face_bboxes, face_landmarks, max_index
if __name__ == '__main__':
det_face_model_path = r'/home/jwq/PycharmProjects/situ/src/face_det/Pytorch_Retinaface/weights/mobilenet_0.25.mnn'
image_path = r'input/3.jpg'
image_save_path = r'results/3.jpg'
thr = 0.5
face_detector = Face_Detector(det_face_model_path)
image = cv2.imread(image_path)
face_detector.detect(image, thr)
# image_ploted = face_detector.plot(image, face_bboxes)
# cv2.imwrite(image_save_path, image_ploted)