Arcface/test_code.py at main · zestyoreo/Arcface · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
import tensorflow as tf
import math
import time
import numpy as np

import hypar

threshold=0.5

def get_embeddings(input_imgs,model):
    test_model = model.resnet
    embeddings = test_model.predict(input_imgs)
    print(embeddings.shape)
    return embeddings

def get_distance(emb1,emb2):
    """
    emb1 & emb2: are both 512 dimensional vectors from the trained resnet model

    get_distance: returns dot_prod,cosine_distance,euclidean_distance
    Check Out "https://github.com/zestyoreo/Arcface/blob/main/get_distance()_test.ipynb" for clarity
    """
    dot_prod = np.dot(emb1,emb2.T)

    a = np.matmul(np.transpose(emb1), emb2)
    b = np.sum(np.multiply(emb1, emb1))
    c = np.sum(np.multiply(emb2, emb2))
    cosine_distance = 1 - (a / (np.sqrt(b) * np.sqrt(c)))

    euclidean_distance = emb1 - emb2
    euclidean_distance = np.sum(np.multiply(euclidean_distance, euclidean_distance))
    euclidean_distance = np.sqrt(euclidean_distance)

    return dot_prod,cosine_distance,euclidean_distance

def recognise(img,face_imgs_database,face_embedding_database,face_names_database):
    """
    img: numpy array of dims (112,112,3) of image of face to be recognised
    face_imgs_database: numpy array of dims (num_of_faces,512)
    face_names_database: dictionary of form {index_of_face:"name_of_person"} comprising of names of faces in face_imgs_database array

    recognise: returns the name and face of the closest match in the database
    """
    img_embedding = get_embeddings(img)
    best_match_index = []
    best_distance = float('-inf')
    for i in range(int(face_embedding_database.shape[0])):
        distance,cosine_distance,euclidean_distance = get_distance(img_embedding,face_embedding_database[i])
        if distance>best_distance:
            best_distance = distance
            best_match_index = [i]
        elif distance==best_distance:
            best_match_index.append(i)

    names = []
    faces = []
    for i in best_match_index:
        names.append(face_names_database[i])
        faces.append(face_imgs_database[i])

    return names,faces

def verify(img,face):
    """
    img: numpy array of dims (112,112,3) of image of face to be verified
    face: image of face of person to be verified with

    verify: returns boolean if the faces match or not
    """
    img_embedding = get_embeddings(img)
    face_embedding = get_embeddings(face)
    distance,cosine_distance,euclidean_distance = get_distance(img_embedding,face_embedding)
    if distance>threshold:
        return True
    return False

def calculate_threshold(X,Y,model):
    index = [None] * X.shape[0]
    embeddings = get_embeddings(X,model)

    same_person_distance = []
    same_person_cosine_distance = []
    same_person_euclidean_distance = []
    for clas in range(hypar.num_classes):
        index[clas] = []
        for i in range (0,X.shape[0]):
            if Y[i] == clas:
                index[clas].append(i)
        for i in range(len(index[clas])):
            i1=index[clas][i]
            for j in range(i+1,len(index[clas])):
                j2=index[clas][j]
                distance,cosine_distance,euclidean_distance = get_distance(embeddings[i1],embeddings[j2])
                same_person_distance.append(distance)
                same_person_cosine_distance.append(cosine_distance)
                same_person_euclidean_distance.append(euclidean_distance)

    same_person_distance = np.asarray(same_person_distance)
    same_person_cosine_distance = np.asarray(same_person_cosine_distance)
    same_person_euclidean_distance = np.asarray(same_person_euclidean_distance)

    same_person_distance_mean = np.mean(same_person_distance)
    same_person_cosine_distance_mean = np.mean(same_person_cosine_distance)
    same_person_euclidean_distance_mean = np.mean(same_person_euclidean_distance)

    diff_person_distance = []
    diff_person_cosine_distance = []
    diff_person_euclidean_distance = []
    for clas in range(hypar.num_classes):
        for c in range(clas+1,hypar.num_classes):
            for i in range(len(index[clas])):
                for j in range(len(index[c])):
                    distance,cosine_distance,euclidean_distance = get_distance(embeddings[i],embeddings[j])
                    diff_person_distance.append(distance)
                    diff_person_cosine_distance.append(cosine_distance)
                    diff_person_euclidean_distance.append(euclidean_distance)

    diff_person_distance = np.asarray(diff_person_distance)
    diff_person_cosine_distance = np.asarray(diff_person_cosine_distance)
    diff_person_euclidean_distance = np.asarray(diff_person_euclidean_distance)

    diff_person_distance_mean = np.mean(diff_person_distance)
    diff_person_cosine_distance_mean = np.mean(diff_person_cosine_distance)
    diff_person_euclidean_distance_mean = np.mean(diff_person_euclidean_distance)

    alpha=0.8
    distance_threshold = diff_person_distance_mean*(1-alpha)+same_person_distance_mean*alpha
    cosine_distance_threshold = diff_person_cosine_distance_mean*(1-alpha)+same_person_cosine_distance_mean*alpha
    euclidean_distance_threshold = diff_person_euclidean_distance_mean*(1-alpha)+same_person_euclidean_distance_mean*alpha
    return distance_threshold, cosine_distance_threshold, euclidean_distance_threshold