-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataloader.py
142 lines (124 loc) · 8.45 KB
/
dataloader.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
import scipy.io as sio
import torch
from sklearn import preprocessing
from sklearn.metrics.pairwise import cosine_similarity
import torch.nn.functional as F
# glob label to local label
def map_label(label, classes):
mapped_label = torch.LongTensor(label.size())
for i in range(classes.size(0)):
mapped_label[label == classes[i]] = i
return mapped_label
# load dataset
class Dataset(object):
def __init__(self, opt):
self.opt = opt
mat = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + opt.image_embedding + ".mat")
self.feature = mat['features'].T
self.label = mat['labels'].astype(int).squeeze() - 1
self.image_file = mat['image_files']
mat = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + opt.class_embedding + "_splits.mat")
trainval_loc = mat['trainval_loc'].squeeze() - 1
test_seen_loc = mat['test_seen_loc'].squeeze() - 1
test_unseen_loc = mat['test_unseen_loc'].squeeze() - 1
self.allclasses = torch.from_numpy(np.unique(self.label))
self.allclasses_num = self.allclasses.size(0)
if opt.dataset == 'CUB':
file_path = opt.dataroot + '/CUB/sent_splits.mat' # 1024D
mat = sio.loadmat(file_path)
self.attribute = F.normalize(torch.from_numpy(mat['att'].T), dim=1).float()
# file_path = opt.dataroot + '/CUB/cub_attributes_reed.npy'
# self.attribute = F.normalize(torch.from_numpy(np.load(file_path)), dim=1).float()
else:
self.attribute = torch.from_numpy(mat['att'].T).float()
if opt.fine_tuning:
if opt.preprocessing:
if opt.standardization:
print('standardization...')
scaler = preprocessing.StandardScaler()
else:
scaler = preprocessing.MinMaxScaler()
_train_feature = scaler.fit_transform(np.load(f'fine_tuning_data/{opt.dataset}/train_feature.npy'))
_test_seen_feature = scaler.fit_transform(np.load(f'fine_tuning_data/{opt.dataset}/test_seen_feature.npy'))
_test_unseen_feature = scaler.fit_transform(np.load(f'fine_tuning_data/{opt.dataset}/test_unseen_feature.npy'))
self.train_feature = torch.from_numpy(_train_feature).float()
mx = self.train_feature.max()
self.train_feature.mul_(1 / mx) # train seen feature
self.test_unseen_feature = torch.from_numpy(_test_unseen_feature).float()
self.test_unseen_feature.mul_(1 / mx) # test unseen feature
self.test_seen_feature = torch.from_numpy(_test_seen_feature).float()
self.test_seen_feature.mul_(1 / mx) # test seen feature
else:
self.train_feature = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/train_feature.npy'))
self.test_seen_feature = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/test_seen_feature.npy'))
self.test_unseen_feature = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/test_unseen_feature.npy'))
self.train_label = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/train_label.npy'))
self.test_seen_label = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/test_seen_label.npy'))
self.test_unseen_label = torch.from_numpy(np.load(f'fine_tuning_data/{opt.dataset}/test_unseen_label.npy'))
else:
if opt.preprocessing:
if opt.standardization:
print('standardization...')
scaler = preprocessing.StandardScaler() # sklearn库中的数据预处理函数（基于numpy）
else:
scaler = preprocessing.MinMaxScaler()
_train_feature = scaler.fit_transform(self.feature[trainval_loc])
_test_seen_feature = scaler.transform(self.feature[test_seen_loc])
_test_unseen_feature = scaler.transform(self.feature[test_unseen_loc])
self.train_feature = torch.from_numpy(_train_feature).float()
mx = self.train_feature.max()
self.train_feature.mul_(1 / mx) # train seen feature
self.test_unseen_feature = torch.from_numpy(_test_unseen_feature).float()
self.test_unseen_feature.mul_(1 / mx) # test unseen feature
self.test_seen_feature = torch.from_numpy(_test_seen_feature).float()
self.test_seen_feature.mul_(1 / mx) # test seen feature
else:
self.train_feature = torch.from_numpy(self.feature[trainval_loc]).float() # train seen feature
self.test_unseen_feature = torch.from_numpy(self.feature[test_unseen_loc]).float() # test unseen feature
self.test_seen_feature = torch.from_numpy(self.feature[test_seen_loc]).float() # test seen feature
self.train_label = torch.from_numpy(self.label[trainval_loc]).long() # train seen label
self.test_seen_label = torch.from_numpy(self.label[test_seen_loc]).long() # test seen label
self.test_unseen_label = torch.from_numpy(self.label[test_unseen_loc]).long() # test unseen label
self.test_unseen_image_file = self.image_file[test_unseen_loc]
self.ntrain = self.train_feature.shape[0]
self.seenclasses = torch.from_numpy(np.unique(self.train_label.numpy())) # seen classes
self.unseenclasses = torch.from_numpy(np.unique(self.test_unseen_label.numpy())) # unseen classes
self.ntest_class = self.unseenclasses.shape[0]
self.train_local_label = map_label(self.train_label, self.seenclasses) # train local label
self.test_seen_local_label = map_label(self.test_seen_label, self.seenclasses) # test seen local label
self.test_unseen_local_label = map_label(self.test_unseen_label, self.unseenclasses) # test unseen local label
self.seenclass_num = self.seenclasses.size(0) # number of seen classes
self.unseenclass_num = self.unseenclasses.size(0) # number of unseen classes
self.seen_att = self.attribute[self.seenclasses] # attribute of seen classes
self.unseen_att = self.attribute[self.unseenclasses] # attribute of unseen classes
self.reset_att = torch.cat((self.seen_att, self.unseen_att), dim=0)
self.feature_dim = self.train_feature.shape[1] # dim of feature
self.att_dim = self.attribute.shape[1] # dim of attribute
self.train_data = [self.train_feature.numpy(), self.train_label.numpy()]
self.test_data = [self.test_unseen_feature.numpy(), self.test_unseen_label.numpy()]
self.class_feature = []
self.class_label = []
self.s_u_semantic_similarity_check(self.opt.seen_Neighbours, self.seen_att, self.unseen_att)
# collect the data of each class
self.tr_cls_centroid = torch.zeros([self.seenclass_num, self.train_feature.shape[1]])
for i in range(self.seenclass_num):
self.tr_cls_centroid[i] = torch.mean(self.train_feature[self.train_local_label == i], dim=0)
self.seen_class_num = []
for i in self.seenclasses:
num = self.train_feature[self.train_label == i].shape[0]
self.seen_class_num.append(num)
self.seen_class_num = np.array(self.seen_class_num)
def s_u_semantic_similarity_check(self, Neighbours, train_text_feature, test_text_feature):
'''
Unseen class
'''
unseen_similarity_matric = torch.from_numpy(cosine_similarity(test_text_feature, train_text_feature))
# Mapping matric
self.s_unseen_idx_mat = torch.argsort(-1 * unseen_similarity_matric, dim=1)
self.s_unseen_idx_mat = self.s_unseen_idx_mat[:, 0:Neighbours]
# Neighbours Semantic similary values
self.s_u_semantic_similarity_unseen = torch.zeros((self.unseenclass_num, Neighbours))
for i in range(self.unseenclass_num):
for j in range(Neighbours):
self.s_u_semantic_similarity_unseen[i, j] = unseen_similarity_matric[i, self.s_unseen_idx_mat[i, j]]