class Age_Gender_Detect():
def __init__(self):
self._detector = MTCNN()
self._fa = FaceAligner()
self._image_utils = Image_Utils()
self._agender = PyAgender()
self._gap_db = Gap_Db()
def detect_face_age_gender(self, fn, debug_mode=False):
print("\nprocessing file %s..." % fn)
# check if faces already detected for the family
family_image_name = os.path.basename(fn).split('.')[0]
family_id = self._gap_db.get_family_id(family_image_name)
if self._gap_db.is_face_done(family_id):
print("Faces already detected for the family! Moving on...\n")
return
else:
# if faces not detected for the family yet, remove previous faces for
# the family
os.system("rm -rf %s" %
(os.path.join(FACE_IMG_DIR, '%s*' % family_image_name)))
img_original = cv2.imread(fn)
if img_original.shape[0] >= img_original.shape[1]:
img = imutils.resize(img_original, height=2000)
else:
img = imutils.resize(img_original, width=2000)
faces = self._detector.detect_faces(img)
self._image_utils.squarify_box(faces)
if debug_mode:
self._image_utils.show_key_points(img, faces)
face_patches = self._image_utils.show_face_patches(img,
faces,
show=False)
# align faces
faces_aligned = self._fa.align(faces, img)
assert len(faces_aligned) == len(face_patches)
# detect age and gender
ages = []
genders = []
for face in face_patches:
g, a = self._agender.gender_age(face)
gender = ['female' if g > 0.5 else 'male'][0]
age = int(a)
ages.append(age)
genders.append(gender)
print("gender: %s, age: %i" % (gender, age))
# save faces
face_images = []
face_no = 0
for face in faces_aligned:
face_fn = family_image_name + '_' + str(face_no) + '.jpg'
face_images.append(face_fn[:-4]) # remove 'extension'
face_f_fn = os.path.join(FACE_IMG_DIR, face_fn)
cv2.imwrite(face_f_fn, face)
face_no += 1
## update database
for image, age, gender in zip(face_images, ages, genders):
self._gap_db.update_faces(family_id, image, age, gender)
for j in range(0,68):
landmarks[i][j] = shape.part(j).x
landmarks[i][j+68] = shape.part(j).y
# expressions
# -----------
frame = cv2.imread(fimg,0)
roi = cv2.resize(frame, (48, 48))
roi = roi.astype("float") / 255.0
roi = img_to_array(roi)
roi = np.expand_dims(roi, axis=0)
expressions = expression_classifier.predict(roi)[0]
if i==0:
m = expressions.shape[0]
data = np.zeros((n,m))
data[i] = expressions
# age and gender
# --------------
gender_age[i] = agender.gender_age(cv2.imread(fimg))
print('saving landmarks.npy, expressions.npy, and gender_age.npy and ... ')
np.save('landmarks',landmarks)
np.save('expressions',data)
np.save('gender_age',gender_age)
