how can i input more picture to improve the accuracy of identification

[ck003]

• face_recognition version:
• Python version:
• Operating System: Raspbian

Description

how can i input more picture to improve the accuracy of identification

[FurqanMehmood]

Add multiple images of same person in known people folder by adding _1,2 ....n at end of name. While adding label in results remove and numbers after _ by using this name=known_names[i] end=name.find('_') name=name[:end]

[Rutulpatel7077]

can you please give more clarification? @FurqanMehmood

[lordcenzin]

I was interested in the same point. What i did is:

• one folder per person, with multiple images, i.e. ./barak_obama/0.jpg, 1.jpg etc.

I used a db to quickly map filepath and name Hope it helps

[FurqanMehmood]

use this tutorial for more details https://www.pyimagesearch.com/2018/06/18/face-recognition-with-opencv-python-and-deep-learning/

[Rutulpatel7077]

@FurqanMehmood Thanks for the link! I haven't tried yet! I will.

If I make the dataset of 1k images of one person. Do you think that might help me to improve accuracy?

[Rutulpatel7077]

@lordcenzin Can you please give some example of how you did implement in code?

[dmastag]

I was looking into this myself and found that there is a KNN example in this repo worth checking out

https://github.com/ageitgey/face_recognition/blob/master/examples/face_recognition_knn.py

[ComputerCarl]

I was wondering if there was a way to do this with the face_recognition library. Perhaps loading n images and then averaging the face_encodings result tensors?

For example, I have 7 photos of Amy. I want to store a (128-scalar) tensor for Amy - not 7 of them. Can I combine them/average them?

[ComputerCarl]

So.. this averaging thing seems to work, but could someone provide an educated opinion?

import face_recognition
import numpy as np
# https://face-recognition.readthedocs.io/en/latest/face_recognition.html

a1 = face_recognition.load_image_file('people/amber1.jpg')
a2 = face_recognition.load_image_file('people/amber2.jpg')
a3 = face_recognition.load_image_file('people/amber3.jpg')

a1_encodings = face_recognition.face_encodings(a1)
a2_encodings = face_recognition.face_encodings(a2)
a3_encodings = face_recognition.face_encodings(a3)

# unknown is amber4
unknown_person = face_recognition.load_image_file('people/unknown.jpg')
unknown_encoding = face_recognition.face_encodings(unknown_person)

two_mean_encodings = np.array((a1_encodings[0] + a2_encodings[0])) / 2
three_mean_encodings = np.array(
    (a1_encodings[0] + a2_encodings[0] + a3_encodings[0])) / 3

distance = face_recognition.face_distance(a1_encodings, unknown_encoding[0])
distance2 = face_recognition.face_distance(
    [two_mean_encodings], unknown_encoding[0])
distance3 = face_recognition.face_distance(
    [three_mean_encodings], unknown_encoding[0])

print('one image', distance)
print('two images', distance2)
print('three images', distance3)

results;
one image [0.38789462]
two images [0.35752516]
three images [0.31829679]

[auveele]

So.. this averaging thing seems to work, but could someone provide an educated opinion?

import face_recognition
import numpy as np
# https://face-recognition.readthedocs.io/en/latest/face_recognition.html

a1 = face_recognition.load_image_file('people/amber1.jpg')
a2 = face_recognition.load_image_file('people/amber2.jpg')
a3 = face_recognition.load_image_file('people/amber3.jpg')

a1_encodings = face_recognition.face_encodings(a1)
a2_encodings = face_recognition.face_encodings(a2)
a3_encodings = face_recognition.face_encodings(a3)

# unknown is amber4
unknown_person = face_recognition.load_image_file('people/unknown.jpg')
unknown_encoding = face_recognition.face_encodings(unknown_person)

two_mean_encodings = np.array((a1_encodings[0] + a2_encodings[0])) / 2
three_mean_encodings = np.array(
    (a1_encodings[0] + a2_encodings[0] + a3_encodings[0])) / 3

distance = face_recognition.face_distance(a1_encodings, unknown_encoding[0])
distance2 = face_recognition.face_distance(
    [two_mean_encodings], unknown_encoding[0])
distance3 = face_recognition.face_distance(
    [three_mean_encodings], unknown_encoding[0])

print('one image', distance)
print('two images', distance2)
print('three images', distance3)

results;
one image [0.38789462]
two images [0.35752516]
three images [0.31829679]

Hi, did you find a solution for this?