-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdist.py
98 lines (91 loc) · 4.76 KB
/
dist.py
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
# -*- coding:utf-8 -*-
import numpy as np
def dist_skilled_forgery(features,template_num,nf,genuine_num,forgery_num):
dist_positive,dist_negative,dist_template = [],[],[]
for i,feat in enumerate(features):
# feat_a = feat[0:template_num]
# feat_p = feat[(template_num + nf):(genuine_num + nf)]
l = len(feat)
first = template_num // 2 # 第一轮拿这么多个，偶数拿一半，奇数拿少一个
second = template_num - first # 第二轮拿这么多个
feat_a = feat[0:first]
feat_a = np.concatenate((feat_a,feat[l // 4:l // 4 + second]),axis=0)
feat_p = np.concatenate((feat[first:l // 4],feat[l // 4 + second:l // 2]),axis=0)
feat_n = feat[(genuine_num + nf):]
dist_p = np.zeros((genuine_num - template_num,template_num))
dist_n = np.zeros((forgery_num,template_num))
dist_t = np.zeros((template_num,template_num))
for i in range(template_num):
fa1 = feat_a[i]
for j in range(i + 1,template_num):
fa2 = feat_a[j]
dist = np.sqrt(np.sum(np.power(fa2 - fa1,2),axis=0))
dist_t[i,j] = dist # 对角线为0，上三角矩阵
for i in range(genuine_num - template_num):
fp = feat_p[i]
for j in range(template_num):
fa = feat_a[j]
dist = np.sqrt(np.sum(np.power(fp - fa,2),axis=0))
dist_p[i,j] = dist
for i in range(forgery_num):
fn = feat_n[i]
for j in range(template_num):
fa = feat_a[j]
dist = np.sqrt(np.sum(np.power(fn - fa,2),axis=0))
dist_n[i,j] = dist
dist_positive.append(dist_p)
dist_negative.append(dist_n)
dist_template.append(dist_t)
dist_positive = np.concatenate(dist_positive,axis=0)
dist_negative = np.concatenate(dist_negative,axis=0)
dist_template = np.concatenate(dist_template,axis=0)
return dist_positive,dist_negative,dist_template
def dist_random_forgery(features,template_num,nf,genuine_num,forgery_num):
dist_positive,dist_negative,dist_template = [],[],[]
features_anchor,features_positive = [],[]
for i,feat in enumerate(features):
# feat_a = feat[0:template_num]
# feat_p = feat[(template_num + nf):(genuine_num + nf)]
l = len(feat)
first = template_num // 2 # 第一轮拿这么多个，偶数拿一半，奇数拿少一个
second = template_num - first # 第二轮拿这么多个
feat_a = feat[0:first]
feat_a = np.concatenate((feat_a,feat[l // 4:l // 4 + second]),axis=0)
feat_p = np.concatenate((feat[first:l // 4],feat[l // 4 + second:l // 2]),axis=0)
features_anchor.append(feat_a)
features_positive.append(feat_p)
for i,feat_a in enumerate(features_anchor): # i是user的下标
feat_p = features_positive[i] # 这里是会变的，主要改变的是这里
feat_n = []
for j in range(len(features_anchor)): # j也是user的下标，一个user有4个anchor16个genuine
if i != j:
feat_n.append(features_positive[j][3]) # 随便选个genuine而已，不能只选一轮的genuine啊
dist_p = np.zeros((len(feat_p),template_num))
dist_n = np.zeros((len(feat_n),template_num))
dist_t = np.zeros((template_num,template_num))
for j in range(template_num):
fa1 = feat_a[j] # (640,)
for k in range(j + 1,template_num):
fa2 = feat_a[k] # anchor与anchor之间
dist = np.sqrt(np.sum(np.power(fa2 - fa1,2),axis=0))
dist_t[j,k] = dist # 对角线为0，上三角矩阵
for j in range(len(feat_p)):
fp = feat_p[j]
for k in range(template_num):
fa = feat_a[k] # anchor与anchor之间
dist = np.sqrt(np.sum(np.power(fp - fa,2),axis=0))
dist_p[j,k] = dist # 每个user都有4个anchor，每个anchor与每个positive的距离
for j in range(len(feat_n)):
fn = feat_n[j] # (16,640)
for k in range(template_num):
fa = feat_a[k] # anchor与negative之间
# dist = np.mean(np.sqrt(np.sum(np.power(fn - np.expand_dims(fa,axis=0),2),axis=1)),axis=0)
dist = np.sqrt(np.sum(np.power(fn - fa,2),axis=0))
dist_n[j,k] = dist # 每个user都有4个anchor，每个anchor与每个positive的距离
dist_positive.append(dist_p)
dist_negative.append(dist_n)
dist_template.append(dist_t)
dist_positive = np.concatenate(dist_positive,axis=0)
dist_negative = np.concatenate(dist_negative,axis=0)
dist_template = np.concatenate(dist_template,axis=0)
return dist_positive,dist_negative,dist_template