我是靠谱客的博主 爱笑书本,这篇文章主要介绍HCP_DataLoader,现在分享给大家,希望可以做个参考。

鉴于数据太大,肯要用S3来存
S3的读取速度应该还是很快的

复制代码
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
from torch.utils.data import Dataset,DataLoader import torch import numpy as np import nibabel as nbl HCPDataListPath = '/home/ec2-user/SageMaker/Models_HCP/dt.txt' HCPDataRootPath = '/home/ec2-user/SageMaker/HCP_dataset/' labelNumber = {'EMOTION': 0, 'GAMBLING': 1, 'LANGUAGE': 2, 'MOTOR': 3, 'RELATIONAL': 4, 'SOCIAL': 5, 'WM': 6} def getLabelList(HCPDataPathList): labelList = [] for HCPDataPath in HCPDataPathList: labelList.append(labelNumber[HCPDataPath.split('_')[1].split('.')[0]]) return np.array(labelList) def getData(HCPDataRootPath, dataPath): dtseries = nbl.load(HCPDataRootPath + dataPath) time_series = dtseries.get_fdata().reshape((-1)) shape = dtseries.header.matrix.get_index_map(1).volume.volume_dimensions nifti = np.zeros(shape) for bm in dtseries.header.matrix.get_index_map(1).brain_models: if bm.model_type == 'CIFTI_MODEL_TYPE_SURFACE': continue voxels = bm.voxel_indices_ijk off, cnt = bm.index_offset, bm.index_count nifti[tuple(np.transpose(voxels))] = time_series[off:off + cnt] return np.array(nifti).reshape((1, 91, 109, 91)).astype(np.float32) class HCPDataSet(DataSet): def __init__(self,HCPDataRootPath,HCPDataPathList,HCPLabelList): self.HCPDataRootPath = HCPDataRootPath self.HCPDataPathList = np.array(HCPDataPathList) self.HCPLabelList = np.array(HCPLabelList) self.total = len(HCPDataPathList) def __getitem__(self,index): dataPath = self.HCPDataPathList[index] data = getData(self.HCPDataRootPath,dataPath) label = np.array(self.HCPLabelList[index]) return torch.from_numpy(data),torch.from_numpy(label) def __len__(self): return self.total class WholeDataSet(): def __init__(self, trainDataSet, evalDataSet, testDataSet): self.trainDataSet = trainDataSet self.evalDataSet = evalDataSet self.testDataSet = testDataSet def getHCPDataSet(HCPDataRootPath,HCPDataListPath,evalRate=0.2,testRate=0.2,tiny_data=0): dataPathList = [] with open(HCPDataListPath,'r') as fr: for HCPDataPath in fr.readlines(): dataPathList.append(HCPDataPath.strip()) if tiny_data != 0: dataPathList = dataPathList[:int(tiny_data)] totalNumber = len(dataPathList) totalTraining = int(totalNumber * (1 - evalRate - testRate)) totalEvaluation = int(totalNumber * evalRate) print('Training : {},Evaluation: {},Test: {}'.format(totalTraining,totalEvaluation,totalNumber-totalTraining-totalEvaluation)) trainDataPathList = dataPathList[:totalTraining] trainLabelList = getLabelList(trainDataPathList) evalDataPathList = dataPathList[totalTraining:totalTraining + totalEvaluation] evalLabelList = getLabelList(evalDataPathList) testDataPathList = dataPathList[totalTraining+totalEvaluation:] testLabelList = getLabelList(testDataPathList) trainDataSet = HCPDataSet(HCPDataRootPath, trainDataPathList, trainLabelList) evalDataSet = HCPDataSet(HCPDataRootPath, evalDataPathList, evalLabelList) testDataSet = HCPDataSet(HCPDataRootPath, testDataPathList, testLabelList) dataSet = WholeDataSet(trainDataSet,evalDataSet,testDataSet) return dataSet dataSet = getHCPDataSet(HCPDataRootPath, HCPDataListPath, evalRate=0.05, testRate=0.15)
复制代码
1
2
3
4
5
6
def getData(HCPDataRootPath, dataPath): data = nbl.load(HCPDataRootPath + dataPath).get_fdata().reshape((-1)) data = np.array(data).reshape((1, 91, 109, 91)).astype(np.float32) data = (data - np.min(data))/np.max(data) - np.min(data) return data

最后

以上就是爱笑书本最近收集整理的关于HCP_DataLoader的全部内容,更多相关HCP_DataLoader内容请搜索靠谱客的其他文章。

本图文内容来源于网友提供,作为学习参考使用,或来自网络收集整理,版权属于原作者所有。
点赞(79)

评论列表共有 0 条评论

立即
投稿
返回
顶部