-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset.py
139 lines (90 loc) · 4.19 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
import os
import torch
from torch.utils import data
from PIL import Image
import numpy as np
from torchvision import transforms as T
import nibabel as nib
import h5py
import random
class create_cross(data.Dataset):
def __init__(self, data_root1, data_root2,transforms=None, train=True, test=False):
self.test = test
datas1 = [os.path.join(data_root1, png) for png in os.listdir(data_root1)]
datas2 = [os.path.join(data_root2, png) for png in os.listdir(data_root2)]
datas1 = sorted(datas1, key=lambda x: (int(x.split('_')[-3].split('/')[-1]),
int(x.split('_')[-1].split('.')[-2])) )
datas2 = sorted(datas2, key=lambda x: (int(x.split('_')[-3].split('/')[-1]),
int(x.split('_')[-1].split('.')[-2])) )
self.train =train
if self.train:
self.datas1 = datas1[int(0.3* len(datas1)):int(1* len(datas1)):3]
self.datas2 = datas2[int(0.3* len(datas2)):int(1* len(datas2)):3]
elif self.test:
self.datas1 = datas1[int(0.9 * len(datas1)):]
self.datas2 = datas2[int(0.9 * len(datas2)):]
else:
self.datas1 = datas1[int(0* len(datas1)):int(0.3 * len(datas1)):2]
self.datas2 = datas2[int(0* len(datas2)):int(0.3 * len(datas2)):2]
self.datas = self.datas1+self.datas2
self.transforms1 = T.RandomRotation(90)
w = np.loadtxt('Label_cross.txt',dtype=int)
self.w = torch.from_numpy(w)
def __getitem__(self, index):
input,label1,label2,label3,label4,Mask1,Mask2,Mask3,Mask4 = self.read(index)
label = torch.cat((label1.unsqueeze(0)+label4.unsqueeze(0),label2.unsqueeze(0)),0).unsqueeze(0).long()
label = (label>0).long()
cross = torch.matmul(label,self.w)
cross = cross.numpy()
if self.train:
if cross.sum() > 0:
index_=random.choice(np.where(cross>0)[1])
else:
index_ = index
else:
index_ = index
input_,label1_,label2_,label3_,label4_,Mask1_,Mask2_,Mask3_,Mask4_ = self.read(index_)
input = torch.cat((input,input_),0)
label1 = torch.cat((label1.unsqueeze(0),label1_.unsqueeze(0)),0)
label2 = torch.cat((label2.unsqueeze(0),label2_.unsqueeze(0)),0)
label4 = torch.cat((label4.unsqueeze(0),label4_.unsqueeze(0)),0)
return input, label1, label2,label4,Mask1+Mask4,Mask1+Mask2+Mask4,self.datas[index]
def read(self,index):
f = h5py.File(self.datas[index],'r')
t1 = np.expand_dims(f['t1'][:],0)
t1ce = np.expand_dims(f['t1ce'][:],0)
t2 = np.expand_dims(f['t2'][:],0)
flair = np.expand_dims(f['flair'][:],0)
gt = np.expand_dims(f['label'][:],0)
t1 = self.normalize(t1)
t1ce = self.normalize(t1ce)
t2 = self.normalize(t2)
flair = self.normalize(flair)
input = np.concatenate((t1,t1ce,t2,flair),axis = 0)
if self.train:
data = np.concatenate((input,gt),axis = 0)
data = torch.from_numpy(data).type(torch.FloatTensor)
data = self.transforms1(data)
input = data[:4,:,:]
gt = data[4,:,:].unsqueeze(0)
else:
input = torch.from_numpy(input).type(torch.FloatTensor)
gt = torch.from_numpy(gt*1.0).type(torch.FloatTensor)
Mask1 = (gt==1).int()
Mask2 = (gt==2).int()
Mask3 = (gt==3).int()
Mask4 = (gt==4).int()
label1 = Mask1.sum()>0
label2 = Mask2.sum()>0
label3 = Mask3.sum()>0
label4 = Mask4.sum()>0
return input,label1,label2,label3,label4,Mask1,Mask2,Mask3,Mask4
def normalize(self, data, smooth=1e-9):
mean = data.mean()
std = data.std()
if (mean == 0) or (std == 0):
return data
data = (data - mean + smooth) / (std + smooth)
return data
def __len__(self):
return len(self.datas)