-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdatasets.py
164 lines (138 loc) · 6.78 KB
/
datasets.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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
import os
import torch
import pandas as pd
from skimage import io
from torch.utils.data import Dataset
import torchvision
import torchvision.transforms as transforms
import numpy as np
from tqdm import tqdm
class DimTransform:
def __init__(self, target_dim, class_split):
self.target_dim = target_dim
self.class_split = class_split
def __call__(self, x):
if np.argmax(x) in self.class_split[0]:
label = torch.zeros(self.target_dim)
label[self.class_split[0].index(np.argmax(x))] = 1
else:
label = torch.ones(self.target_dim)
label = -1. / self.target_dim * label
return label
class Subset(Dataset):
def __init__(self, data, label, data_split, transform=None, label_transform=None):
self.data = data
self.label = label
self.data_split = data_split
self.transform = transform
self.label_transform = label_transform
def __getitem__(self, index):
data = self.data[index]
label = self.label[index]
if self.transform is not None:
data = self.transform(data)
if self.label_transform is not None:
label = self.label_transform(label)
return data, label
def __len__(self):
return len(self.data)
class Base(Dataset):
def __init__(self, img_dir, label_dir, transform, aug_transform, label_transform):
self.data_dir = img_dir
self.label_dir = label_dir
self.transform = transform
self.aug_transform = aug_transform
self.label_transform = label_transform
self.data = []
self.label = []
self.load_data()
self.idx_by_class = []
self.train_idx = []
self.valid_idx_id = []
self.valid_idx_ood = []
def load_data(self):
pass
def __getitem__(self, index: int) -> (np.ndarray, np.ndarray):
data = self.data[index]
label = self.label[index]
if self.transform is not None:
data = self.transform(data)
if self.label_transform is not None:
label = self.label_transform(label)
return data, label
def __len__(self) -> int:
return len(self.data)
def split(self, fold: int, class_split: tuple,
data_transform: torchvision.transforms = None,
aug_transform: torchvision.transforms = None,
split_label_transform: torchvision.transforms = None) -> (Subset, Subset, Subset):
assert len(class_split) == 2, f'Wrong split setting is given! expect 2, given {len(class_split)}.'
if data_transform is None:
data_transform = self.transform
if aug_transform is None:
aug_transform = self.aug_transform
if split_label_transform is None:
split_label_transform = DimTransform(len(class_split[0]), class_split=class_split)
# split data according to categories
for i in range(self.label.shape[-1]):
idx = np.where(np.argmax(self.label, axis=1) == i)[0]
self.idx_by_class.append(idx)
for class_id in class_split[1]:
self.valid_idx_ood += self.idx_by_class[class_id].tolist()
idx_id = np.setdiff1d(np.arange(len(self.data)), self.valid_idx_ood)
# valid_idx = np.linspace(fold, len(idx_id), len(idx_id) // 5, endpoint=False, dtype=np.int)
self.valid_idx_id = idx_id[np.linspace(fold, len(idx_id), len(idx_id) // 5, endpoint=False, dtype=np.int)]
self.train_idx = np.setdiff1d(idx_id, self.valid_idx_id)
# self.train_idx = self.train_idx[np.linspace(0, len(self.train_idx), len(self.train_idx) // 5, endpoint=False, dtype=np.int)]
train_set = Subset([self.data[idx] for idx in self.train_idx], [self.label[idx] for idx in self.train_idx],
data_split=class_split,
transform=aug_transform,
label_transform=split_label_transform)
valid_set_id = Subset([self.data[idx] for idx in self.valid_idx_id],
[self.label[idx] for idx in self.valid_idx_id],
data_split=class_split,
transform=data_transform,
label_transform=split_label_transform)
valid_set_ood = Subset([self.data[idx] for idx in self.valid_idx_ood],
[self.label[idx] for idx in self.valid_idx_ood],
data_split=class_split,
transform=data_transform,
label_transform=split_label_transform)
return train_set, valid_set_id, valid_set_ood
class ISIC(Base):
def __init__(self, img_dir, label_dir, transform=None, aug_transform=None, label_transform=None):
super(ISIC, self).__init__(img_dir, label_dir, transform, aug_transform, label_transform)
if transform is None:
self.transform = transforms.Compose([
transforms.ToPILImage(), transforms.Resize((224, 224)), transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])
if aug_transform is None:
self.aug_transform = transforms.Compose([
transforms.ToPILImage(), transforms.Resize((224, 224)), transforms.ToTensor(),
transforms.RandomCrop((224, 224), padding=4),
transforms.RandomHorizontalFlip(), transforms.RandomVerticalFlip(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])
def load_data(self):
if not os.path.exists(self.data_dir):
raise RuntimeError('data path does not exist!')
if not os.path.exists(self.label_dir):
raise RuntimeError('label path does not exist!')
if os.path.isfile(self.label_dir): # 读取csv标签
csv_reader = pd.read_csv(self.label_dir, header=0, index_col='image')
else:
raise RuntimeError('wrong label path is given!')
print(f'Start loading ISIC from {self.data_dir}')
with tqdm(total=len(os.listdir(self.data_dir)), ncols=100) as _tqdm:
for step, img in enumerate(os.listdir(self.data_dir)):
p_img = os.path.join(self.data_dir, img)
if p_img.endswith('jpg'):
data = io.imread(p_img)
id = img.split('.')[0]
label = np.array(csv_reader.loc[id])
self.data.append(data)
self.label.append(label)
_tqdm.update(1)
self.label = np.array(self.label)
print('Finish loading data!')