def run_frame(self, frame, fcp):
age_result = {}
gender_result = {}
frame_original = frame.copy()
(h, w, c) = frame.shape
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
cfg_size = (608, 608) # keep same as net input
frame_input = cv2.resize(frame_rgb, cfg_size)
threshold = 0.24
dets = libpysunergy.detect(frame_input.data, w, h, c, threshold,
self.net, self.names)
for i in range(len(dets)):
if dets[i][4] > 0 and (dets[i][5] - dets[i][4]) > 40:
[fleft, fright, ftop, fbot] = dets[i][2:6]
face_img = frame_original[ftop:fbot, fleft:fright].copy()
(fh, fw, fc) = face_img.shape
#face recognition
face_image = cv2.resize(face_img, (112, 96))
face_image = face_image[:, :, ::-1].transpose((2, 0, 1))
face_image = (face_image[np.newaxis, :, :, :] - 127.5) / 128.0
face_image = torch.from_numpy(face_image).float()
face_image = Variable(face_image).cuda()
output = self.frnet(face_image).data[0].tolist()
ret, faceid = fcp.match(output)
if ret:
age, gender = faceid.split(':')[0], faceid.split(':')[1]
if not ret:
#end of face recognition
dets2 = libpysunergy.predict(face_img.data, fw, fh, fc,
self.top, self.net2,
self.names2)
age = dets2[0][0]
dets3 = libpysunergy.predict(face_img.data, fw, fh, fc,
self.top, self.net3,
self.names3)
gender = dets3[0][0]
age, gender = res_conv(int(age), gender)
fcp.insert(output, str(age) + ":" + str(gender))
if age not in age_result:
age_result[age] = 1
else:
age_result[age] += 1
if gender not in gender_result:
gender_result[gender] = 1
else:
gender_result[gender] += 1
return age_result, gender_result
def run_frame_det(self, frame, frame_num):
result = []
frame_original = frame.copy()
(h, w, c) = frame.shape
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
cfg_size = (608, 608) # keep same as net input
frame_input = cv2.resize(frame_rgb, cfg_size)
threshold = 0.24
dets = libpysunergy.detect(frame_input.data, w, h, c, threshold,
self.net, self.names)
for i in range(len(dets)):
if dets[i][4] > 0 and (dets[i][5] - dets[i][4]) > 40:
[fleft, fright, ftop, fbot] = dets[i][2:6]
#in the format of MOT17
detection = {}
detection['score'] = float(dets[i][1])
detection['bbox'] = (float(fleft), float(ftop), float(fright),
float(fbot))
result.append(detection)
return result
def predict(ori, dest):
avi = dest.split('.')[0] + '.avi'
threshold = 0.24
#load video
cap = cv2.VideoCapture(ori)
fps = cap.get(cv2.cv.CV_CAP_PROP_FPS)
width = int(cap.get(cv2.cv.CV_CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.cv.CV_CAP_PROP_FRAME_HEIGHT))
print fps
#load face detector
net1, names = libpysunergy.load('data/face.data', 'cfg/yolo-face.cfg',
'weights/yolo-face.weights')
#load AGE models
net2, names2 = libpysunergy.load("data/age.data", "cfg/age.cfg",
"weights/age.weights")
net3, names3 = libpysunergy.load("data/gender.data", "cfg/gender.cfg",
"weights/gender.weights")
net4, names4 = libpysunergy.load("data/race.data", "cfg/race.cfg",
"weights/race.weights")
top = 1
# Define the codec and create VideoWriter object
fourcc = cv2.cv.CV_FOURCC('M', 'J', 'P', 'G')
videoWriter = cv2.VideoWriter(avi, fourcc, fps, (width, height))
font = ImageFont.truetype("Roboto-Regular.ttf", 20)
count = 1
#face detection
while (1):
print count
count += 1
ret, frame = cap.read()
if not ret:
break
cv2_im = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
pil_im = Image.fromarray(cv2_im)
draw = ImageDraw.Draw(pil_im)
if ret == True:
(h, w, c) = frame.shape
dets = libpysunergy.detect(frame.data, w, h, c, threshold, net1,
names)
#crop face and predict AGE
for i in range(0, len(dets)):
if dets[i][0] == 'face':
box = dets[i][2:6]
x0 = int(box[0])
x1 = int(box[1])
y0 = int(box[2])
y1 = int(box[3])
faceimg = frame[y0:y1, x0:x1].copy()
(h, w, c) = faceimg.shape
#draw bounding box
draw.rectangle(
((x0, y0), (x1, y1)),
outline="red",
)
draw.rectangle(
((x0 + 1, y0 + 1), (x1 - 1, y1 - 1)),
outline="red",
)
dets2 = libpysunergy.predict(faceimg.data, w, h, c, top,
net2, names2)
age = dets2[0][0]
dets3 = libpysunergy.predict(faceimg.data, w, h, c, top,
net3, names3)
gender = dets3[0][0]
dets4 = libpysunergy.predict(faceimg.data, w, h, c, top,
net4, names4)
race = dets4[0][0]
#write classification
draw.text((x0, y0 - 60),
'Age: ' + age, (255, 0, 0),
font=font)
draw.text((x0, y0 - 40),
'Gender: ' + gender, (255, 0, 0),
font=font)
draw.text((x0, y0 - 20),
'Race: ' + race, (255, 0, 0),
font=font)
pil_im = cv2.cvtColor(np.array(pil_im), cv2.COLOR_RGB2BGR)
videoWriter.write(pil_im)
else:
break
cap.release()
libpysunergy.free(net1)
libpysunergy.free(net2)
libpysunergy.free(net3)
libpysunergy.free(net4)
#convert avi video to mp4
print('converting avi video to mp4 ...')
cm.getstatusoutput('ffmpeg -i ' + avi + ' -c:v libx264 -crf ' +
str(int(fps)) +
' -preset slow -c:a libfdk_aac -b:a 192k -ac 2 ' + dest)
cm.getstatusoutput('rm ' + avi)
def run_frame_visual(self, frame, frame_num, fcp):
fff = open('result/result.txt', 'w')
dic = {}
'''
while 1:
line = fff.readline()
if not line:
break
id = line.split(' ')[0] + '-'+line.split(' ')[1]
dic[id] = line.split(' ')[2]+'_'+line.split(' ')[3]
'''
age_result = {}
gender_result = {}
frame_original = frame.copy()
(h, w, c) = frame.shape
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
cfg_size = (608, 608) # keep same as net input
frame_input = cv2.resize(frame_rgb, cfg_size)
threshold = 0.24
dets = libpysunergy.detect(frame_input.data, w, h, c, threshold,
self.net, self.names)
dets_num = len(dets)
fmin = 9999
fmax = -1
for i in range(len(dets)):
if dets[i][4] > 0 and (dets[i][5] - dets[i][4]) > 40 and (
dets[i][3] - dets[i][2]) > 20 and dets[i][5] < 150:
[fleft, fright, ftop, fbot] = dets[i][2:6]
if fleft < fmin:
fmin = fleft
if fleft > fmax:
fmax = fleft
for i in range(len(dets)):
if dets[i][4] > 0 and (dets[i][5] - dets[i][4]) > 40 and (
dets[i][3] - dets[i][2]) > 20 and dets[i][5] < 150:
[fleft, fright, ftop, fbot] = dets[i][2:6]
face_img = frame_original[ftop:fbot, fleft:fright].copy()
(fh, fw, fc) = face_img.shape
#face recognition
face_image = cv2.resize(face_img, (112, 96))
face_image = face_image[:, :, ::-1].transpose((2, 0, 1))
face_image = (face_image[np.newaxis, :, :, :] - 127.5) / 128.0
face_image = torch.from_numpy(face_image).float()
face_image = Variable(face_image).cuda()
output = self.frnet(face_image).data[0].tolist()
ret, faceid = fcp.match(output)
if ret:
age, gender = faceid.split(':')[0], faceid.split(':')[1]
if not ret:
#end of face recognition
dets2 = libpysunergy.predict(face_img.data, fw, fh, fc,
self.top, self.net2,
self.names2)
age = dets2[0][0]
dets3 = libpysunergy.predict(face_img.data, fw, fh, fc,
self.top, self.net3,
self.names3)
gender = dets3[0][0]
age, gender = res_conv(int(age), gender)
fcp.insert(output, str(age) + ":" + str(gender))
if fleft == fmin:
age = '30-35'
gender = 'Female'
if fleft == fmax:
age = '30-35'
gender = 'Male'
fff.write(
str(frame_num) + ' ' + str(fleft) + ' ' + age + ' ' +
gender + '\n')
'''
if str(frame_num) +'-'+str(fleft) in dic:
age = dic[str(frame_num)+'-'+str(fleft)].split('_')[0]
gender = dic[str(frame_num)+'-'+str(fleft)].split('_')[1][:-1]
else:
continue
'''
cv2.rectangle(frame, (fleft, ftop), (fright, fbot),
(0, 0, 255), 2)
frame = cv2.putText(frame, 'Age:' + age, (fleft, ftop),
cv2.FONT_HERSHEY_COMPLEX, 0.5,
(255, 255, 255), 1)
frame = cv2.putText(frame, 'Gender:' + gender,
(fleft, ftop + 20),
cv2.FONT_HERSHEY_COMPLEX, 0.5,
(255, 255, 255), 1)
return frame
def predict(net0, net1, net2, names0, names1, names2, frame):
age_frame, gender_frame = static_init()
try:
(img_h, img_w, img_c) = frame.shape
img_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
cfg_size = (1024, 1024)
img_input = cv2.resize(img_rgb, cfg_size)
faces = libpysunergy.detect(img_input.data, img_w, img_h, img_c,
threshold, net0, names0)
facenum = len(faces)
#{py_names[class], prob, left, right, top, bot}
for i in range(0, facenum):
[x0, x1, y0,
y1] = [faces[i][2], faces[i][3], faces[i][4], faces[i][5]]
x0 = max(x0, 0)
y0 = max(y0, 0)
x1 = min(x1, img_w - 1)
y1 = min(y1, img_h - 1)
faceimg = img_rgb[y0:y1, x0:x1].copy()
(h, w, c) = faceimg.shape
#get result from sunergy
dets1 = libpysunergy.predict(faceimg.data, w, h, c, top, net1,
names1)
age = int(str(dets1[0][0]))
dets2 = libpysunergy.predict(faceimg.data, w, h, c, top, net2,
names2)
gender = dets2[0][0]
if gender == 'male':
gender_frame['male'] += 1
elif gender == 'female':
gender_frame['female'] += 1
if int(age) < 20:
age_frame['<20'] += 1
if int(age) >= 20 and int(age) < 25:
age_frame['20-25'] += 1
if int(age) >= 25 and int(age) < 30:
age_frame['25-30'] += 1
if int(age) >= 30 and int(age) < 35:
age_frame['30-35'] += 1
if int(age) >= 35 and int(age) < 40:
age_frame['35-40'] += 1
if int(age) >= 40 and int(age) < 45:
age_frame['40-45'] += 1
if int(age) >= 45 and int(age) < 50:
age_frame['45-50'] += 1
if int(age) >= 50 and int(age) < 55:
age_frame['50-55'] += 1
if int(age) >= 55 and int(age) < 60:
age_frame['55-60'] += 1
if int(age) >= 60:
age_frame['>60'] += 1
return age_frame, gender_frame
except Exception, e:
print('Error on line {}'.format(sys.exc_info()[-1].tb_lineno),
type(e).__name__, e)