-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
143 lines (128 loc) · 7.14 KB
/
dataset.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
143
import os
import torch
import numpy as np
import torch.nn as nn
from torch import Tensor
import torch.nn.functional as F
from torchvision import datasets
from torchvision import transforms
from torch.utils.data.sampler import Sampler
from torch.utils.data import Dataset, DataLoader
from typing import Type, Any, Callable, Union, List, Optional, cast, Tuple
from utils import *
def lab_conv(knownclass, label):
knownclass = sorted(knownclass)
label_convert = torch.zeros(len(label), dtype=int)
for j in range(len(label)):
for i in range(len(knownclass)):
if label[j] == knownclass[i]:
label_convert[j] = int(knownclass.index(knownclass[i]))
break
else:
label_convert[j] = int(len(knownclass))
return label_convert
## for open set
def get_dataset_splits(image_size,batch_size, dataset, selected_known_classes, ratio, is_open):
augmentations = [transforms.RandomHorizontalFlip(p=0.5),transforms.RandomVerticalFlip(p=0.5),
#transforms.ColorJitter(brightness=0.2,contrast=0.2,saturation=0.1,hue=0.1),
transforms.RandomRotation(degrees=(-180, 180)),transforms.RandomAffine(degrees=0, translate=(0.1, 0.1)),
transforms.Resize(image_size),transforms.ToTensor()]
train_transform = transforms.Compose(augmentations)
test_transform = transforms.Compose([transforms.Resize(image_size),transforms.ToTensor()])
if dataset == 'skin':
train_path = os.path.join('./data/ISIC_2018/' , 'train')
test_path = os.path.join('./data/ISIC_2018/' , 'test')
train_class_num = np.array([327,514,1099,115,1113,6705,142])
test_class_num = np.array([43,93,217,44,171,909,35])
elif dataset == 'pathology':
train_path = os.path.join('./data/NCT-CRC-HE-100K/' , 'train')
test_path = os.path.join('./data/NCT-CRC-HE-100K/' , 'test')
train_class_num = np.array([7285,7396,8058,8090,6227,9476,6134,7312,10022])
test_class_num = np.array([3122,3170,3454,3467,2669,4060,2629,3134,4295])
elif dataset == 'DDR':
train_path = os.path.join('./data/DDRCls/' , 'train')
test_path = os.path.join('./data/DDRCls/' , 'test')
train_class_num = np.array([2992,1638,2238,613,2370,575])
test_class_num = np.array([1880,189,1344,71,275,346])
train_dataset = datasets.ImageFolder(train_path, train_transform, loader=pil_loader)
train_cumu_sum = np.cumsum(train_class_num).astype(int)
train_cumu_sum = np.insert(train_cumu_sum, 0, 0, axis=0)
train_label_num = (train_class_num*ratio).astype(int)
label_samples,unlabel_samples = [], []
for i in range(len(train_class_num)):
if i in selected_known_classes:
label_samples = label_samples+train_dataset.samples[train_cumu_sum[i]:train_cumu_sum[i]+train_label_num[i]]
unlabel_samples = unlabel_samples+train_dataset.samples[train_cumu_sum[i]+train_label_num[i]:train_cumu_sum[i+1]]
elif is_open:
unlabel_samples = unlabel_samples+train_dataset.samples[train_cumu_sum[i]:train_cumu_sum[i+1]]
train_dataset.samples = label_samples + unlabel_samples
labeled_bs = int(batch_size/2)
print("=> Total samples: {}, labeled is: {}".format(len(train_dataset), len(label_samples)))
labeled_idxs = list(range(0, len(label_samples)))*int(len(unlabel_samples)/len(label_samples))
unlabeled_idxs = list(range(len(label_samples), len(train_dataset)))
batch_sampler = TwoStreamBatchSampler(labeled_idxs, unlabeled_idxs, batch_size, batch_size-labeled_bs)
ID_test_dataset = datasets.ImageFolder(test_path, test_transform, loader=pil_loader)
test_cumu_sum = np.cumsum(test_class_num).astype(int)
test_cumu_sum = np.insert(test_cumu_sum, 0, 0, axis=0)
test_samples = []
for i in range(len(test_class_num)):
if i in selected_known_classes:
test_samples = test_samples+ID_test_dataset.samples[test_cumu_sum[i]:test_cumu_sum[i+1]]
ID_test_dataset.samples = test_samples
train_targets,test_targets = [], []
for x, y in train_dataset.samples:
train_targets = train_targets+[y]
for x, y in ID_test_dataset.samples:
test_targets = test_targets+[y]
train_trans_labels = lab_conv(selected_known_classes, train_targets)
test_trans_labels = lab_conv(selected_known_classes, test_targets)
for i in range(len(train_trans_labels)):
x,y = train_dataset.samples[i]
train_dataset.samples[i]=(x,train_trans_labels[i])
for i in range(len(test_trans_labels)):
x,y = ID_test_dataset.samples[i]
ID_test_dataset.samples[i]=(x,test_trans_labels[i])
OOD_test_dataset = datasets.ImageFolder(test_path, test_transform, loader=pil_loader)
OOD_trans_labels = lab_conv(selected_known_classes, OOD_test_dataset.targets)
OOD_samples = []
for i in range(len(OOD_trans_labels)):
x,y = OOD_test_dataset.samples[i]
OOD_test_dataset.samples[i]=(x,OOD_trans_labels[i])
print_class_info(train_dataset)
print_class_info(ID_test_dataset)
print_class_info(OOD_test_dataset)
return batch_sampler,train_dataset, ID_test_dataset, OOD_test_dataset
## for close set
def generate_dataset(dataset,input_size,ratio = 0.6):
augmentations = [transforms.RandomHorizontalFlip(p=0.5),transforms.RandomVerticalFlip(p=0.5),
transforms.ColorJitter(brightness=0.2,contrast=0.2,saturation=0.1,hue=0.1),
transforms.RandomRotation(degrees=(-180, 180)),
transforms.RandomAffine(degrees=0, translate=(0.1, 0.1)),
transforms.Resize((input_size, input_size)),transforms.ToTensor()]
train_transform = transforms.Compose(augmentations)
test_transform = transforms.Compose([transforms.Resize((input_size, input_size)),transforms.ToTensor()])
if dataset == 'skin':
train_path = os.path.join('./data/ISIC_2018/' , 'train')
test_path = os.path.join('./data/ISIC_2018/' , 'test')
train_class_num = np.array([327,514,1099,115,1113,6705,142])
elif dataset == 'pathology':
train_path = os.path.join('./data/NCT-CRC-HE-100K/' , 'train')
test_path = os.path.join('./data/NCT-CRC-HE-100K/' , 'test')
train_class_num = np.array([7285,7396,8058,8090,6227,9476,6134,7312,10022])
elif dataset == 'DDR':
train_path = os.path.join('./data/DDRCls/' , 'train')
test_path = os.path.join('./data/DDRCls/' , 'test')
train_class_num = np.array([2992,1638,2238,613,2370,575])
train_dataset = datasets.ImageFolder(train_path, train_transform, loader=pil_loader)
print_class_info(train_dataset)
cumu_sum = np.cumsum(train_class_num).astype(int)
label_num = (train_class_num*ratio).astype(int)
samples = train_dataset.samples[0:label_num[0]]
for i in range(len(train_class_num)):
samples = samples+train_dataset.samples[cumu_sum[i]:cumu_sum[i]+label_num[i]]
train_dataset.samples = samples
test_dataset = datasets.ImageFolder(test_path, test_transform, loader=pil_loader)
print_class_info(test_dataset)
dataset = train_dataset,test_dataset,test_dataset
print_dataset_info(dataset)
return dataset