Why encoding a face with the "num_jitters" parameter affects the encoding of another face (with the "num_jitters" parameter)

user_17759412 0
Viewed 144

Why encoding a face with the "num_jitters" parameter affects the encoding of another face (with the "num_jitters" parameter as well)?

I have an face_encode variable, which is "Har.png" image encode (with num_jitters = 10)

When processing the same image and comparing them, I get the same value (face_encode and test are identical).

However, after uncommenting this line: new = encode("Ani.png", 10)[0] "Face_encode" and "test" become non-identical (np.allclose(face_encode, test) return False). This means that processing one face affects the encoding of another face.

Note: If we will change the num_itters to 1 in new = encode("Ani.png", 1)[0], the print(np.allclose(face_encode, test)) will give a True. This mean that jittering affect on face encode of another image

import cv2
import face_recognition
import numpy as np


face_encode = np.array([-0.03383258,  0.1522091 ,  0.00778023, -0.02392919, -0.12134701,
        0.02448359, -0.12119899,  0.00130317,  0.17482917, -0.00608639,
        0.07468938,  0.01706824, -0.21361005,  0.04690374, -0.10125306,
        0.13845484, -0.23481546, -0.1003552 , -0.09002965, -0.08902221,
        0.009046  ,  0.06367773, -0.01593469,  0.04676583, -0.17100905,
       -0.31411591, -0.08375659, -0.10543557,  0.04868974, -0.12692517,
        0.02808667, -0.0059777 , -0.1129951 , -0.03772227,  0.02010387,
        0.03789502, -0.04583764, -0.06205031,  0.18120724,  0.11010844,
       -0.12902826,  0.04678808,  0.03990007,  0.37248898,  0.17482039,
        0.10718967,  0.02596617, -0.05392049,  0.14223413, -0.26670438,
        0.07981005,  0.22440179,  0.02152029,  0.08013226,  0.06920523,
       -0.12548962,  0.0365652 ,  0.18923151, -0.24735497,  0.07125427,
        0.07109451,  0.0370005 , -0.02891098, -0.07094251,  0.22352193,
        0.13761494, -0.1292235 , -0.11606612,  0.12919654, -0.16925484,
       -0.08498519, -0.04782813, -0.11302817, -0.13826026, -0.34820935,
        0.0340745 ,  0.40744266,  0.17069298, -0.10228713,  0.01735803,
       -0.04485941, -0.05013431,  0.01530067,  0.05060982, -0.19970088,
       -0.01855213, -0.04743322,  0.05434962,  0.15309754,  0.0527404 ,
       -0.02121131,  0.14636354,  0.03430861, -0.0306342 ,  0.01154729,
        0.04730092, -0.2055707 , -0.0561384 , -0.15399964, -0.05845002,
        0.04407276, -0.04798348,  0.02185482,  0.16238688, -0.17582887,
        0.19120233,  0.0242596 , -0.08013647, -0.03035241,  0.07667085,
       -0.17627513, -0.04133784,  0.25034377, -0.17164004,  0.16634598,
        0.18308365,  0.08592145,  0.08905195,  0.11517368,  0.04827499,
       -0.08297287,  0.03901137, -0.15337615, -0.08358597,  0.01340009,
       -0.05940268,  0.05974644,  0.03913205])

def encode(image, jit):
    #Declare you :)
    known_image = cv2.imread(image)
    reversed_img =  cv2.cvtColor(known_image, cv2.COLOR_BGR2RGB)
    face_location = face_recognition.face_locations(reversed_img)
    your_face_encoding = face_recognition.face_encodings(reversed_img, face_location, num_jitters = jit)
    return your_face_encoding

# new = encode("Ani.png", 10)[0]
test = encode("Har.png", 10)[0]


print(np.allclose(face_encode, test))

You can test it with any image.. And u can print the test value (and compare with person_encode ) to make sure that the value (encode) is changed after computing the new encode.

face-recognition python
1 Answers

Check if you are understanding right what is num_jetters is and what that "jittering" is doing (hint: jittering is changing your image slightly and RANDOMLY each cycle and makes face encoding for that changed image. And than returns some average value from all encodings). By default num_jitters = 1 , so setting it in 0 is senseless. Finally, when some A = B and after some manupalation A <> B, it doesn't mean that both A and B are changed, it's enough the only one to be changed, right?

user_151711330
Related