-
Notifications
You must be signed in to change notification settings - Fork 5
/
make_gt_cub.py
142 lines (115 loc) · 6.37 KB
/
make_gt_cub.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
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
131
132
133
134
135
136
137
138
139
140
141
import numpy as np
from scipy._lib._util import _asarray_validated
from scipy.stats import entropy
def compute_ClsSimilarity_underPart(class_attributes, part_type):
ClsSimilarity_underPart = np.zeros((200, 200))
for i_p in range(part_type):
part_attributes = softmax(class_attributes[i_p], axis=1)
for i in range(200):
for j in range(i+1, 200):
ClsSimilarity_underPart[i, j] = ClsSimilarity_underPart[i, j] + Dominik2003IT(part_attributes[i], part_attributes[j])
ClsSimilarity_underPart[j, i] = ClsSimilarity_underPart[i, j]
return ClsSimilarity_underPart / part_type
def Dominik2003IT(distribution1, distribution2):
term1 = distribution1 * np.log((2.0 * distribution1) / (distribution1 + distribution2))
term2 = distribution2 * np.log((2.0 * distribution2) / (distribution1 + distribution2))
return np.sum(term1 + term2)
def logsumexp(a, axis=None, b=None, keepdims=False, return_sign=False):
a = _asarray_validated(a, check_finite=False)
if b is not None:
a, b = np.broadcast_arrays(a, b)
if np.any(b == 0):
a = a + 0. # promote to at least float
a[b == 0] = -np.inf
a_max = np.amax(a, axis=axis, keepdims=True)
if a_max.ndim > 0:
a_max[~np.isfinite(a_max)] = 0
elif not np.isfinite(a_max):
a_max = 0
if b is not None:
b = np.asarray(b)
tmp = b * np.exp(a - a_max)
else:
tmp = np.exp(a - a_max)
# suppress warnings about log of zero
with np.errstate(divide='ignore'):
s = np.sum(tmp, axis=axis, keepdims=keepdims)
if return_sign:
sgn = np.sign(s)
s *= sgn # /= makes more sense but we need zero -> zero
out = np.log(s)
if not keepdims:
a_max = np.squeeze(a_max, axis=axis)
out += a_max
if return_sign:
return out, sgn
else:
return out
def softmax(x, axis=None):
# compute in log space for numerical stability
return np.exp(x - logsumexp(x, axis=axis, keepdims=True))
flist = './cub200/CUB_200_2011/attributes/class_attribute_labels_continuous.txt'
part_num = 15
class_num = 200
attributes = np.zeros((200, 312))
class_index = 0
with open(flist, 'r') as rf:
for line in rf.readlines():
attributes_per_class = line.strip().split()
attributes[class_index, :] = np.float32(attributes_per_class)
class_index = class_index + 1
similarityMatrix_cls_part = np.zeros((class_num, class_num, part_num))
for i_part in range(part_num):
if i_part == 0:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 59 - 1:73], attributes[:, 237 - 1: 240]], 2)
if i_part == 1:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 1 - 1:9], attributes[:, 150 - 1: 152], attributes[:, 279 - 1:293]], 3)
if i_part == 2:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 198 - 1:212], attributes[:, 245 - 1: 248]], 2)
if i_part == 3:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 55 - 1:58], attributes[:, 106 - 1: 120]], 2)
if i_part == 4:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 95 - 1:105], attributes[:, 153 - 1: 167], attributes[:, 294 - 1: 308]], 3)
if i_part == 5:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 95 - 1:105], attributes[:, 153 - 1: 167]], 2)
if i_part == 6:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 136 - 1:149], attributes[:, 95 - 1: 105]], 2)
if i_part == 7:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 264 - 1:278]], 1)
if i_part == 8:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 10 - 1:24], attributes[:, 213 - 1: 217], attributes[:, 309 - 1: 312]], 3)
if i_part == 9:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 183 - 1:197]], 1)
if i_part == 10:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 136 - 1:149], attributes[:, 95 - 1: 105]], 2)
if i_part == 11:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 264 - 1:278]], 1)
if i_part == 12:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 10 - 1:24], attributes[:, 213 - 1: 217], attributes[:, 309 - 1: 312]], 3)
if i_part == 13:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 74 - 1:79], attributes[:, 80 - 1: 94], attributes[:, 168 - 1: 182], attributes[:, 241 - 1: 244]], 4)
if i_part == 14:
similarityMatrix_cls_part[:, :, i_part] = compute_ClsSimilarity_underPart([attributes[:, 95 - 1:105], attributes[:, 121 - 1: 135]], 2)
np.save('./cub200/similarityMatrix_cls_part.npy', similarityMatrix_cls_part)
similarityMatrix_cls_part = np.load('./cub200/similarityMatrix_cls_part.npy')
np.save('./cub200/Dominik2003IT_similarityMatrix_cls_part.npy', similarityMatrix_cls_part)
similarityMatrix_cls_part = np.load('./cub200/Dominik2003IT_similarityMatrix_cls_part.npy')
# threshold and remain most unsimilar and distinguising parts between each class pair
threshold = np.sort(similarityMatrix_cls_part.flatten())[int(0.2*class_num*class_num*part_num)] # the less, the more similar
DistinctMatrix_cls_part_copy = np.copy(similarityMatrix_cls_part)
DistinctMatrix_cls_part_copy[similarityMatrix_cls_part <= threshold] = 0
DistinctMatrix_cls_part_copy[similarityMatrix_cls_part > threshold] = 1
DistinctMatrix_cls_part_copy = DistinctMatrix_cls_part_copy.astype(int)
distinct_extracted_attributes = np.zeros((class_num, class_num), dtype=object)
num_tmp = 0
for i in range(class_num):
for j in range(i+1, class_num):
part_idx = np.argwhere(DistinctMatrix_cls_part_copy[i, j, :] == 1).flatten()
part_idx = part_idx.tolist()
if len(part_idx) == 0:
print(i,j)
num_tmp = num_tmp + 1
distinct_extracted_attributes[i, j] = part_idx
distinct_extracted_attributes[j, i] = part_idx
np.save('./cub200/Dominik2003IT_dis_extracted_attributes_02.npy', distinct_extracted_attributes)
print(num_tmp)