OpenCV 4 TypeError: Expected cv::UMat for argument 'labels'

Tyler Strouth picture Tyler Strouth · Jan 20, 2019 · Viewed 37k times · Source

I am writing a facial recognition program and I keep getting this error when I try to train my recognizer

TypeError: Expected cv::UMat for argument 'labels'

my code is

def detect_face(img):
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")
    faces = face_cascade.detectMultiScale(gray, scaleFactor=1.2, minNeighbors=5);
    if (len(faces)==0):
        return None, None
    (x, y, w, h) = faces[0]
    return gray[y:y+w, x:x+h], faces[0]

def prepare_training_data():
    faces = []
    labels = []
    for img in photo_name_list: #a collection of file locations as strings
        image = cv2.imread(img)
        face, rect = detect_face(image)
        if face is not None:
            faces.append(face)
            labels.append("me")
    return faces, labels

def test_photos():
    face_recognizer = cv2.face.LBPHFaceRecognizer_create()
    faces, labels = prepare_training_data()
    face_recognizer.train(faces, np.ndarray(labels))

labels is list of labels for each photo in the image list returned from prepare_training_data, and I convert it to a numpy array because I read that is what train() needs it to be.

Answer

PolyGlot picture PolyGlot · Jan 21, 2019

Solution - labels should be list of integers, and you should use numpy.array(labels) (or np.array(labels)).

Dummy example to check an error absence:

labels=[0]*len(faces)
face_recognizer.train(faces, np.array(labels))

I haven't found any documentation for openCV face recognizers on python, so I've started to look over c++ documentation and examples. And due to documentation this library uses labels input for train as a std::vector<int>. A cpp example, provided by openCV docs, also uses vector<int> labels. And so on, library even have an error for not an integer input.