class AgePredictor:
def __init__(self):
# age model
# model structure: https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/age.prototxt
# pre-trained weights: https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/dex_chalearn_iccv2015.caffemodel
self.age_model = cv2.dnn.readNetFromCaffe(
"data/age.prototxt", "data/dex_chalearn_iccv2015.caffemodel")
self.fd = FaceDetector()
# given an image
# extract roi using face detector and predict the age using the age model
# 3 return values:
## apparent_age is the model predicted age of the face
## roi is the region of the image that contains the face
## angle is the rotation angle (in CCW) for the roi
def predict_age(self, img):
# extract roi and resize it to the desired dimensions for the age model
roi, angle = self.fd.detect_face(img)
if roi is None:
return -1, None, 0
roi_resized = cv2.resize(roi, (224, 224))
img_blob = cv2.dnn.blobFromImage(roi_resized)
# run it through the model and return predicted age
self.age_model.setInput(img_blob)
age_dist = self.age_model.forward()[0]
output_indexes = np.array([i for i in range(0, 101)])
apparent_age = round(np.sum(age_dist * output_indexes), 2)
return apparent_age, roi, angle
class FaceDetectorProcessor(ImageProcessor): """FaceDetector processing class extending the Image processing base class, attempts to determine if there is a face in the image Parameters: ----------- cascade_file: path to the xml cascade classifier parameters """ def __init__(self, cascade_file='haarcascade_frontalface_default.xml'): self.detector = FaceDetector(cascade_file) self.preprocessor = GrayscaleProcessor() def process_image(self, image): """Process the image by determining if there is a face :param image: image as numpy array :return: cropped face image, and boolean indicating whether there is a face """ gray = self.preprocessor.process_image(image) face = self.detector.detect_face(gray) if len(face) == 0: return image, False x, y, w, h = face cropped_face = image[y:y+h, x:x+w] return cropped_face, True def save_image(self, image, user_id, photo_id): """Saves the image to a temporary directory in the current working folder with a concatentation of user_id and photo_id as the filename :param image: image as a numpy array :param user_id: user_id of the face image :param photo_id: photo_id of the original instagram image has .jpg postfix :return: file path name """ path = os.path.dirname(__file__) path = os.path.join(path, 'tmp') if not os.path.exists(path): os.mkdir(path) fname = os.path.join(path, str(user_id) + str(photo_id)) cv2.imwrite(fname, image) return fname
class FaceDetectorProcessor(ImageProcessor):
"""FaceDetector processing class extending the Image processing
base class, attempts to determine if there is a face in the image
Parameters:
-----------
cascade_file: path to the xml cascade classifier parameters
"""
def __init__(self, cascade_file='haarcascade_frontalface_default.xml'):
self.detector = FaceDetector(cascade_file)
self.preprocessor = GrayscaleProcessor()
def process_image(self, image):
"""Process the image by determining if there is a face
:param image: image as numpy array
:return: cropped face image, and boolean indicating whether
there is a face
"""
gray = self.preprocessor.process_image(image)
face = self.detector.detect_face(gray)
if len(face) == 0:
return image, False
x, y, w, h = face
cropped_face = image[y:y + h, x:x + w]
return cropped_face, True
def save_image(self, image, user_id, photo_id):
"""Saves the image to a temporary directory in the current working
folder with a concatentation of user_id and photo_id as the filename
:param image: image as a numpy array
:param user_id: user_id of the face image
:param photo_id: photo_id of the original instagram image has .jpg postfix
:return: file path name
"""
path = os.path.dirname(__file__)
path = os.path.join(path, 'tmp')
if not os.path.exists(path):
os.mkdir(path)
fname = os.path.join(path, str(user_id) + str(photo_id))
cv2.imwrite(fname, image)
return fname
bottom = min(top + new_width,img_height)
right = min(left+new_width,img_width)
return BBox(int(left),int(right),int(top),int(bottom))
# self.emotions で指定した7次元のベクトルを推定値として返す
#
# input
# input_tensor 入力テンソル
# output
# 推定値ベクトル
def _tensor2emotion(self, input_tensor):
ret_emotions = []
#print("_tensor2emotion input_tensor size",len(input_tensor))
preds = self.recognizer.predict(input_tensor)
#print("_tensor2emotion preds size",len(preds))
return preds
if __name__ == "__main__":
output_dir="../data/outputs"
face_detect_conf_dir="../conf/face-detect"
erecognizer_conf_dir="../conf/emotion-recognizer"
image_path="../data/inputs/meeting_11_304.jpg"
face_detector = FaceDetector(face_detect_conf_dir)
box_faces = face_detector.detect_face(image_path)
emotion_recognizer = EmotionRecognizer(erecognizer_conf_dir, 416, output_dir)
ret_emotions = emotion_recognizer.classify_emotion(image_path, box_faces)
emotion_recognizer.put_caption(image_path, box_faces, ret_emotions)