-
Notifications
You must be signed in to change notification settings - Fork 3
/
faceTestUsingTorch.py
319 lines (298 loc) · 14.8 KB
/
faceTestUsingTorch.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
import cv2
from PIL import Image
import torch
import torchvision.transforms as transforms
from torchTrain import Net
import pandas as pd
import torch.nn as nn
import torch.nn.functional as F
from torchvision import models
net = models.alexnet(pretrained=True)
# use single cell (256*144) to predict not use captured face
# def recognize(classes, frame, namedict, frameCounter, net_path, studyCollection, time_slot, viewInfo):
# # print('view format: ', viewInfo)
# row = viewInfo.get('Row')
# column = viewInfo.get('Column')
# clip_width = int(viewInfo.get('Width') / row)
# clip_height = int(viewInfo.get('Height') / column)
#
# grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# for j in range(row):
# for i in range(column):
# image = frame[clip_height * i:clip_height * (i + 1), clip_width * j:clip_width * (j + 1)]
# # opencv to PIL: BGR2RGB
# PIL_image = cv2pil(image)
# if PIL_image is None:
# continue
# # using model to recognize
# label = predict_model(PIL_image, net_path, len(classes))
#
# # cv2.rectangle(frame, (x - 10, y - 10), (x + w + 10, y + h + 10), (0, 0, 255), 1)
# cv2.putText(frame, classes[label], (clip_width * j+20, clip_height * i+20), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
# print('i: ', i, 'j: ', j, classes[label])
# if frameCounter % time_slot == 1: # every one time slot reset
# for k in studyCollection.keys():
# studyCollection[k] = 0
#
# if not namedict[classes[label]]:
# namedict[classes[label]].append(frameCounter)
# namedict[classes[label]].append(1)
# else:
# namedict[classes[label]][1] += 1
#
# # get the time of this student appear in a time slot
# studyCollection[classes[label]] += 1
#
#
# return frame, namedict, studyCollection
# use every captured face to predict in one cell, if one frame detect and predict successfully, then the left 24 frames (1s has 25 frames) not detect and predict
# use time 14min for 5min test video in windows, 12min in mac
def recognize(classes, frame, namedict, frameCounter, net_path, studyCollection, viewInfo, tmp_dict):
classfier = cv2.CascadeClassifier('./haarcascades/haarcascade_frontalface_default.xml')
grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
row = viewInfo.get('Row')
column = viewInfo.get('Column')
clip_width = int(viewInfo.get('Width') / row) # 256
clip_height = int(viewInfo.get('Height') / column) # 144
fps = viewInfo.get('fps')
recognize_period = viewInfo.get('recognize_period')
study_period = viewInfo.get('study_period')
if frameCounter % int(fps * recognize_period) == 0: # every recognize period reset tmp_dict
tmp_dict.clear()
print('clear tmp dict')
if frameCounter % int(fps * study_period) == 1: # every study period reset
for k in studyCollection.keys():
studyCollection[k] = 0
label = -1
try:
for j in range(row):
for i in range(column):
if (str(j), str(i)) in tmp_dict.keys():
label = tmp_dict[str(j), str(i)]
cv2.putText(frame, classes[label], (clip_width * j +30, clip_height * i+30),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
else:
cropped = grey[clip_height * i:clip_height * (i + 1), clip_width * j:clip_width * (j +1)] # single cell
# cv2.imshow("cropped", cropped)
face_rects = classfier.detectMultiScale(cropped, scaleFactor=1.2, minNeighbors=3, minSize=(32, 32))
# print('num of detected face: ', len(face_rects))
# print([j, i])
# cv2.waitKey(200)
if len(face_rects) > 0:
for face_rect in face_rects:
x, y, w, h = face_rect
image = cropped[y - 10:y + h + 10, x - 10:x + w + 10]
# opencv to PIL: BGR2RGB
PIL_image = cv2pil(image)
if PIL_image is None:
continue
# using model to recognize
label = predict_model(PIL_image, net_path, classes)
if label != -1:
# cv2.rectangle(frame, (x - 10, y - 10), (x + w + 10, y + h + 10), (0, 0, 255), 1)
cv2.putText(frame, classes[label], (clip_width * j +30, clip_height * i +30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
tmp_dict[(str(j), str(i))] = label
else:
continue
if label != -1:
if namedict[classes[label]]==[]:
namedict[classes[label]].append(frameCounter)
namedict[classes[label]].append(1)
else:
namedict[classes[label]][1] += 1
# get the time of this student appear in a time slot
studyCollection[classes[label]] += 1
label = -1
except Exception as e:
print("frame number:", frameCounter, e)
pass
return frame, namedict, studyCollection, tmp_dict
# use every captured face to predict in one frame not one cell, if one frame detect and predict successfully, then the left 24 frames (1s has 25 frames) not detect and predict
# use time 27min for 5min test video
# def recognize(classes, frame, namedict, frameCounter, net_path, studyCollection, time_slot, viewInfo, tmp_dict):
# classfier = cv2.CascadeClassifier('./haarcascades/haarcascade_frontalface_default.xml')
#
# grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# row = viewInfo.get('Row')
# column = viewInfo.get('Column')
# clip_width = int(viewInfo.get('Width') / row) # 256
# clip_height = int(viewInfo.get('Height') / column) # 144
#
# if frameCounter % int(time_slot / 20) == 0: # every second reset tmp_dict
# tmp_dict.clear()
# print('clear tmp dict')
# for j in range(row):
# for i in range(column):
# if (str(j), str(i)) in tmp_dict.keys():
# historical_name = tmp_dict[str(j), str(i)]
# cv2.putText(frame, historical_name, (clip_width * j +30, clip_height * i+30),
# cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
#
# face_rects = classfier.detectMultiScale(grey, scaleFactor=1.2, minNeighbors=3, minSize=(32, 32))
# if len(face_rects) > 0:
# for face_rect in face_rects:
# x, y, w, h = face_rect
# for j in range(row):
# for i in range(column):
# if clip_width*j <= x <= clip_width*(j + 1) and clip_height * i <= y <= clip_height * (i + 1) and (str(j), str(i)) not in tmp_dict.keys():
# image = grey[y - 10:y + h + 10, x - 10:x + w + 10]
# # opencv to PIL: BGR2RGB
# PIL_image = cv2pil(image)
# if PIL_image is None:
# continue
# # using model to recognize
# label = predict_model(PIL_image, net_path, len(classes))
# # cv2.rectangle(frame, (x - 10, y - 10), (x + w + 10, y + h + 10), (0, 0, 255), 1)
# cv2.putText(frame, classes[label], (clip_width * j +30, clip_height * i +30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
# tmp_dict[(str(j), str(i))] = classes[label]
#
# if frameCounter % time_slot == 1: # every one time slot reset
# for k in studyCollection.keys():
# studyCollection[k] = 0
#
# if not namedict[classes[label]]:
# namedict[classes[label]].append(frameCounter)
# namedict[classes[label]].append(1)
# else:
# namedict[classes[label]][1] += 1
#
# # get the time of this student appear in a time slot
# studyCollection[classes[label]] += 1
#
#
# return frame, namedict, studyCollection, tmp_dict
# use captured face to predict in one frame not one cell, if one frame detect and predict successfully, then the left 24 frames (1s has 25 frames) detect but not predict
# use timein 27mfor 5min test video
# def recognize(classes, frame, namedict, frameCounter, net_path, studyCollection, time_slot, viewInfo, tmp_dict):
# classfier = cv2.CascadeClassifier('./haarcascades/haarcascade_frontalface_default.xml')
#
# grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# row = viewInfo.get('Row')
# column = viewInfo.get('Column')
# clip_width = int(viewInfo.get('Width') / row) # 256
# clip_height = int(viewInfo.get('Height') / column) # 144
# switch = 1
# if frameCounter % int(time_slot / 20) == 0: # every second reset tmp_dict
# tmp_dict.clear()
# print('clear tmp dict')
# face_rects = classfier.detectMultiScale(grey, scaleFactor=1.2, minNeighbors=3, minSize=(32, 32))
# if len(face_rects) > 0:
# for face_rect in face_rects:
# x, y, w, h = face_rect
# for i in range(column):
# for j in range(row):
# if clip_width*j <= x <= clip_width*(j + 1) and clip_height * i <= y <= clip_height * (i + 1):
# if (str(j), str(i)) in tmp_dict.keys():
# historical_name = tmp_dict[str(j), str(i)]
# cv2.putText(frame, historical_name, (clip_width * j + 30, clip_height * i + 30),
# cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
# switch = 0
# break
# else:
# image = frame[y - 10:y + h + 10, x - 10:x + w + 10]
# # opencv to PIL: BGR2RGB
# PIL_image = cv2pil(image)
# if PIL_image is None:
# continue
# # using model to recognize
# label = predict_model(PIL_image, net_path, len(classes))
# cv2.putText(frame, classes[label], (clip_width * j +30, clip_height * i +30), cv2.FONT_HERSHEY_SIMPLEX, 1,
# (255, 0, 0), 2, cv2.LINE_AA) # label name
#
# tmp_dict[(str(j), str(i))] = classes[label]
#
# if frameCounter % time_slot == 1: # every one time slot reset
# for k in studyCollection.keys():
# studyCollection[k] = 0
#
# if not namedict[classes[label]]:
# namedict[classes[label]].append(frameCounter)
# namedict[classes[label]].append(1)
# else:
# namedict[classes[label]][1] += 1
#
# # get the time of this student appear in a time slot
# studyCollection[classes[label]] += 1
# if switch == 0:
# break
# if switch == 0:
# switch = 1
# continue
#
# return frame, namedict, studyCollection, tmp_dict
# use captured face to predict, and detect and predict every frame
# use time 42min for 5min test video
# def recognize(classes, frame, namedict, frameCounter, net_path, studyCollection, time_slot):
# classfier = cv2.CascadeClassifier('./haarcascades/haarcascade_frontalface_default.xml')
#
# grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# face_rects = classfier.detectMultiScale(grey, scaleFactor=1.2, minNeighbors=3, minSize=(32, 32))
# if len(face_rects) > 0:
#
# for face_rect in face_rects:
# x, y, w, h = face_rect
# image = frame[y - 10:y + h + 10, x - 10:x + w + 10]
# # opencv to PIL: BGR2RGB
# PIL_image = cv2pil(image)
# if PIL_image is None:
# continue
# # using model to recognize
# label = predict_model(PIL_image, net_path, len(classes))
# cv2.rectangle(frame, (x - 10, y - 10), (x + w + 10, y + h + 10), (0, 0, 255), 1)
# cv2.putText(frame, classes[label], (x - 10, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2, cv2.LINE_AA) # label name
#
# if frameCounter % time_slot == 1: # every one time slot reset
# for k in studyCollection.keys():
# studyCollection[k] = 0
#
# if not namedict[classes[label]]:
# namedict[classes[label]].append(frameCounter)
# namedict[classes[label]].append(1)
# else:
# namedict[classes[label]][1] += 1
#
# # get the time of this student appear in a time slot
# studyCollection[classes[label]] += 1
#
#
# return frame, namedict, studyCollection
def get_transform():
return transforms.Compose([
transforms.Resize((224, 224)), # reszie image to 224*224
transforms.CenterCrop(224), # center crop 224*224
transforms.ToTensor() # each pixel to tensor
])
def cv2pil(image):
if image.size != 0:
return Image.fromarray(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
else:
return None
def predict_model(image, net_path, classes):
data_transform = get_transform()
image = data_transform(image) # change PIL image to tensor
image = image.view(-1, 3, 224, 224)
# net = Net(class_num)
net.classifier = nn.Sequential(
nn.Dropout(),
nn.Linear(256 * 6 * 6, 4096),
nn.ReLU(inplace=True),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, len(classes)),
)
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
net.to(DEVICE)
# load net
net.load_state_dict(torch.load(net_path))
output = net(image.to(DEVICE))
prob = F.softmax(output[0], dim=0).detach()
idx = torch.argmax(prob).item()
# pred = output.max(1, keepdim=True)[1]
# if pred.item() != idx:
# print('no', pred.item())
# print('output:', prob)
if prob[idx] > 0.95:
return idx
else:
return -1