【分割数据集操作集锦】毕设记录
1. 按要求将CSV文件转成json文件
有时候一些网络模型的源码会有data.json这样的文件里面存放了训练集和验证集的信息，这里我们根据csv格式的表格生成json文件。
以下代码有下述功能：
- 选出同时出现在csv文件里和训练集文件夹中同名的文件按照自己需要加入列表。
- 将序号补全成文件名 fname='amos_'+str(id).zfill(4)+'.nii.gz'
- 根据json格式将1中列表补入字典
# 读取数据
import os
domainAB=[]
domainC=[]
imglist = os.listdir('/media/fsk/DATA1/AMOS22_total/imagesTr')
import csv
with open('/home/fsk/monai/nnprocess/data_perpare/data/data1.csv',encoding='UTF-8-sig') as csvfile:
reader=csv.DictReader(csvfile)
for i,row in enumerate(reader):
id=row['id']
fname='amos_'+str(id).zfill(4)+'.nii.gz'
#print(fname)
if fname in imglist:
if i>=0 and i<=249:
domainAB.append(fname)
if i>=372 and i<=499:
domainC.append(fname)
# print(domainAB)
# print(domainC)
dataset={
"name": "AMOS",
"description": "Amos: A large-scale abdominal multi-organ benchmark for versatile medical image segmentation",
"author": "Yuanfeng Ji",
"reference": "SRIDB x CUHKSZ x HKU x LGCHSZ x LGPHSZ",
"licence": "CC-BY-SA 4.0",
"release": "1.0 01/05/2022",
"contact": "[email protected]",
"tensorImageSize": "3D",
"modality": {"0": "CT"},
"labels": {
"0": "background",
"1": "spleen",
"2": "gall bladder",
"3": "esophagus",
"4": "liver",
"5": "stomach",
"6": "arota",
"7": "pancreas",
"8": "right adrenal gland",
"9": "left adrenal gland"
},
"numTraining": len(domainAB)+len(domainC),
"numTest":len(domainC)
}
datasetDG={
"name": "AMOS",
"description": "Amos: A large-scale abdominal multi-organ benchmark for versatile medical image segmentation",
"author": "Yuanfeng Ji",
"reference": "SRIDB x CUHKSZ x HKU x LGCHSZ x LGPHSZ",
"licence": "CC-BY-SA 4.0",
"release": "1.0 01/05/2022",
"contact": "[email protected]",
"tensorImageSize": "3D",
"modality": {"0": "CT"},
"labels": {
"0": "background",
"1": "spleen",
"2": "gall bladder",
"3": "esophagus",
"4": "liver",
"5": "stomach",
"6": "arota",
"7": "pancreas",
"8": "right adrenal gland",
"9": "left adrenal gland"
},
"numTraining": len(domainAB),
"numTest":len(domainC)
}
training=[]
trainingDG=[]
test=[]
for i in range(len(domainAB)):
img="./imagesTr/"+domainAB[i]
label="./labelsTr/"+domainAB[i]
dic={"image":img,"label":label}
training.append(dic)
trainingDG.append(dic)
for i in range(len(domainC)):
img="./imagesTr/"+domainC[i]
label="./labelsTr/"+domainC[i]
dic={"image":img,"label":label}
training.append(dic)
test.append(img)
dataset['training']=training
datasetDG['training']=trainingDG
dataset['test']=test
datasetDG['test']=test
import json
with open('data/dataset.json','w') as fp:
json.dump(dataset,fp)
with open('data/datasetDG.json','w') as fp:
json.dump(datasetDG,fp)
2. 批量复制文件，将dir1中和dir2中交集文件全复制到dir3
# 导入os模块和shutil模块
import os
import shutil
# 定义三个文件夹的路径
dir1 = "/media/fsk/DATA1/AMOS22_total/labelsTr"
dir2 = "/media/fsk/DATA1/nnunet/nnUNet_raw/nnUNet_raw_data/Task216_AMOS2022_task2_AB/inferTs"
dir3 = "/media/fsk/DATA1/nnunet/nnUNet_raw/nnUNet_raw_data/Task216_AMOS2022_task2_AB/labelsTs"
# 遍历dir1中的文件
for file in os.listdir(dir1):
# 拼接文件的完整路径
file_path = os.path.join(dir1, file)
# 判断是否是文件，而不是文件夹
if os.path.isfile(file_path):
# 判断dir2中是否存在同名文件
if os.path.exists(os.path.join(dir2, file)):
# 复制文件到dir3中，如果已存在则覆盖
shutil.copy(file_path, dir3)3. 批量修改文件名
- 补全文件名，比如在末尾加_000
- 将"img"改成"label"
import os
folder_path="/media/fsk/DATA1/BTCV/imagesTr"
for file in os.listdir(folder_path):
filepath=os.path.join(folder_path,file)
newfile=file.split('.')[0]+"_0000.nii.gz"
#newfile=file.replace("img","label")
newpath=os.path.join(folder_path,newfile)
os.rename(filepath,newpath)4. 多个数据集标签统一
这边需要特别注意！！！！
在替换标签值的时候注意顺序，比如： 如果先将label=1的设为label=5 然后再将label=5的设为label=7，那么label=1和label=5的都会变成label=7。
#顺序
#"0": "background", "1": "spleen", "2": "gall bladder", "3": "esophagus", "4": "liver", "5": "stomach", "6": "arota", "7": "pancreas", "8": "right adrenal gland", "9": "left adrenal gland"
# 导入nibabel包
import nibabel as nib
import numpy as np
import os
import time
# 读取amos数据和标签
input_path='/media/fsk/DATA1/AbdomenCT/Mask'
output_path='/media/fsk/DATA1/AbdomenCT/new_Mask'
labels = os.listdir(input_path)
for label in labels:
# print(label,os.path.join(input_path,label))
amos_label = nib.load(os.path.join(input_path,label))
# 获取浮点数矩阵
amos_flabel = amos_label.get_fdata()
# 获取不同的标签值
print(label,np.unique(amos_flabel))
# 替换你想要修改的标签值
amos_flabel[np.where(amos_flabel == 6)] = 0
amos_flabel[np.where(amos_flabel == 2)] = 0
amos_flabel[np.where(amos_flabel == 12)] = 0
amos_flabel[np.where(amos_flabel == 5)] = 6
amos_flabel[np.where(amos_flabel == 7)] = 5
amos_flabel[np.where(amos_flabel == 4)] = 7
amos_flabel[np.where(amos_flabel == 1)] = 4
amos_flabel[np.where(amos_flabel == 3)] = 1
amos_flabel[np.where(amos_flabel == 8)] = 2
amos_flabel[np.where(amos_flabel == 9)] = 3
amos_flabel[np.where(amos_flabel == 10)] = 8
amos_flabel[np.where(amos_flabel == 11)] = 9
new_amos_label=nib.Nifti1Image(amos_flabel,amos_label.affine)
nib.save(new_amos_label,os.path.join(output_path,label))
new_label = nib.load(os.path.join(output_path,label))
# 获取浮点数矩阵
new_flabel = amos_label.get_fdata()
# 获取不同的标签值
print(label,np.unique(new_flabel))5. 读取pkl数据
#读取.pkl格式的文件
import pickle
path='/media/fsk/DATA1/nnunet/nnUNet_preprocessed/Task216_AMOS2022_task1/nnUNetPlans_bfnnUNet_fabresnet_31_plans_3D.pkl'
f=open(path,'rb')
data=pickle.load(f)
print(data)
print(len(data))