-
Notifications
You must be signed in to change notification settings - Fork 10
/
mislabel_cifar.py
142 lines (123 loc) · 5.23 KB
/
mislabel_cifar.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
142
import torch
import torchvision
from PIL import Image
import torchvision.transforms as transforms
import numpy as np
import copy
import warnings
class MISLABELCIFAR10(torchvision.datasets.CIFAR10):
def __init__(self, root, mislabel_type='agnostic', mislabel_ratio=0.5, rand_number=0, train=True,
transform=None, target_transform=None,
download=False):
super(MISLABELCIFAR10, self).__init__(root, train, transform, target_transform, download)
np.random.seed(rand_number)
self.gen_mislabeled_data(mislabel_type=mislabel_type, mislabel_ratio=mislabel_ratio)
def gen_mislabeled_data(self, mislabel_type, mislabel_ratio):
"""Gen a list of imbalanced training data, and replace the origin data with the generated ones."""
new_targets = []
num_cls = np.max(self.targets) + 1
if mislabel_type == 'agnostic':
for i, target in enumerate(self.targets):
if np.random.rand() < mislabel_ratio:
new_target = target
while new_target == target:
new_target = np.random.randint(num_cls)
new_targets.append(new_target)
else:
new_targets.append(target)
elif mislabel_type == 'asym':
ordered_list = np.arange(num_cls)
while True:
permu_list = np.random.permutation(num_cls)
if np.any(ordered_list == permu_list):
continue
else:
break
for i, target in enumerate(self.targets):
if np.random.rand() < mislabel_ratio:
new_target = permu_list[target]
new_targets.append(new_target)
else:
new_targets.append(target)
else:
warnings.warn('Noise type is not listed')
self.real_targets = self.targets
self.targets = new_targets
self.whole_data = self.data.copy()
self.whole_targets = copy.deepcopy(self.targets)
self.whole_real_targets = copy.deepcopy(self.real_targets)
def switch_data(self):
self.data = self.whole_data
self.targets = self.whole_targets
self.real_targets = self.whole_real_targets
def adjust_base_indx_tmp(self, idx):
new_data = self.whole_data[idx, ...]
targets_np = np.array(self.whole_targets)
new_targets = targets_np[idx].tolist()
real_targets_np = np.array(self.whole_real_targets)
new_real_targets = real_targets_np[idx].tolist()
self.data = new_data
self.targets = new_targets
self.real_targets = new_real_targets
def adjust_base_indx_perma(self, idx):
new_data = self.whole_data[idx, ...]
targets_np = np.array(self.whole_targets)
new_targets = targets_np[idx].tolist()
real_targets_np = np.array(self.whole_real_targets)
new_real_targets = real_targets_np[idx].tolist()
self.whole_data = new_data
self.whole_targets = new_targets
self.whole_real_targets = new_real_targets
self.data = self.whole_data
self.targets = self.whole_targets
self.real_targets = self.whole_real_targets
def estimate_label_acc(self):
targets_np = np.array(self.targets)
real_targets_np = np.array(self.real_targets)
label_acc = np.sum((targets_np == real_targets_np)) / len(targets_np)
return label_acc
def fetch(self, targets):
whole_targets_np = np.array(self.whole_targets)
uniq_targets = np.unique(whole_targets_np)
idx_dict = {}
for uniq_target in uniq_targets:
idx_dict[uniq_target] = np.where(whole_targets_np == uniq_target)[0]
idx_list = []
for target in targets:
idx_list.append(np.random.choice(idx_dict[target.item()], 1))
idx_list = np.array(idx_list).flatten()
imgs = []
for idx in idx_list:
img = self.whole_data[idx]
img = Image.fromarray(img)
img = self.transform(img)
imgs.append(img[None, ...])
train_data = torch.cat(imgs, dim=0)
return train_data
def __getitem__(self, index):
img, target, real_target = self.data[index], self.targets[index], self.real_targets[index]
img = Image.fromarray(img)
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target, real_target, index
class MISLABELCIFAR100(MISLABELCIFAR10):
"""`CIFAR100 <https://www.cs.toronto.edu/~kriz/cifar.html>`_ Dataset.
This is a subclass of the `CIFAR10` Dataset.
"""
base_folder = 'cifar-100-python'
url = "https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz"
filename = "cifar-100-python.tar.gz"
tgz_md5 = 'eb9058c3a382ffc7106e4002c42a8d85'
train_list = [
['train', '16019d7e3df5f24257cddd939b257f8d'],
]
test_list = [
['test', 'f0ef6b0ae62326f3e7ffdfab6717acfc'],
]
meta = {
'filename': 'meta',
'key': 'fine_label_names',
'md5': '7973b15100ade9c7d40fb424638fde48',
}