Face alignment is working correctly?

[ojrandom1]

Is this a bug or a wrong usage of deepface?

The aligned image (img_aligned) does not show up as aligned in

from deepface import DeepFace
from deepface.detectors import FaceDetector
import cv2

detector_name = 'mtcnn'
detector = FaceDetector.build_model('mtcnn')

img = cv2.imread("tests/dataset/img31.jpg")
img_aligned = cv2.imread("tests/dataset/aligned.jpg")
try:
    faces = FaceDetector.detect_faces(detector, detector_name, img, align=True)
except:
    print("error")

for face, (x, y, w, h) in faces:
    region = [x, y, w, h]
    custom_face = img[y:y+h, x:x+w]
    cv2.imwrite("test/aligned.jpg", custom_face)

[serengil]

could you share the aligned version here?

[serengil]

Try this:

img = DeepFace.detectFace("tests/dataset/img31.jpg")
cv2.imwrite("base_image.jpg", img*255)

[ojrandom1]

Yes,

detector_name = 'mtcnn' detector = FaceDetector.build_model('mtcnn')

img = DeepFace.detectFace("tests/dataset/img31.jpg") cv2.imwrite("base_image.jpg", img*255)

works as expected and produces an aligned image, see

What is the recommended solution to have have aligned faces for the face representations for example using an approach like this.

    detector_name = 'mtcnn'
    detector = FaceDetector.build_model('mtcnn')

    model = DeepFace.build_model('ArcFace')

    img = cv2.imread(path)
    try:
        faces = FaceDetector.detect_faces(detector, detector_name, img, align=True)
    except:
        print("error")

    for face, (x, y, w, h) in faces:
        if w < 100
            continue
        region = [x, y, w, h]
        custom_face = img[y:y+h, x:x+w]

        input_shape = functions.find_input_shape(model)
        input_shape_x = input_shape[0]
        input_shape_y = input_shape[1]

        custom_face = functions.preprocess_face(
            img=custom_face,
            target_size=(input_shape_y, input_shape_x),
            enforce_detection=False,
            detector_backend=detector_name,
            align = True)
        representation = []
        if custom_face.shape[1:3] == input_shape:
            representation = self.model.predict(custom_face)[0, :]
        else:
            continue