生成monai框架的dataset的json文件

前言

在使用monai.data.CacheDataset导入数据，以及monai.data.DataLoader时，使用json文件组织数据。

那么假设我们的数据图像文件放在images文件夹里面，标签数据放在labels文件夹里面，如何生成这种格式的json文件呢，并且训练集和验证集的比例为8:2

2023/05/05更新

import json
import os

"""
数据组织结构：

data_base_dir:
    images_dir:
        train
        val
    labels_dir:
        train
        val
"""


if __name__ == '__main__':
    file_dict = {}

    images_base_dir = "/home/data/hablee_data_dir/aorta2023/images"
    labels_base_dir = "/home/data/hablee_data_dir/aorta2023/mergeFourLabel"

    train_images_dir = os.path.join(images_base_dir, "train")
    train_labels_dir = os.path.join(labels_base_dir, "train")

    val_images_dir = os.path.join(images_base_dir, "val")
    val_labels_dir = os.path.join(labels_base_dir, "val")

    train_filenames = os.listdir(train_labels_dir)
    val_filenames = os.listdir(val_labels_dir)

    # -> train
    train_content = []
    for filename in train_filenames:
        image_path = os.path.join(train_images_dir, filename)
        label_path = os.path.join(train_labels_dir, filename)

        dict_content = {"image": image_path, "label":label_path}
        train_content.append(dict_content)

    # -> val
    val_content = []
    for filename in val_filenames:
        image_path = os.path.join(val_images_dir, filename)
        label_path = os.path.join(val_labels_dir, filename)

        dict_content = {"image": image_path, "label":label_path}
        val_content.append(dict_content)


    labels_contents = {"0":"background", "1":"fdm", "2": "zdm", "3":"xxg", "4":"fei" }
    file_dict["labels"] = labels_contents

    file_dict["train"] = train_content
    file_dict["val"] = val_content

    # -> save
    with open("four_cls.json", "w") as f:
        json.dump(file_dict, f, indent=4)
        

代码

import json
import os
import random

if __name__ == '__main__':
    demo = False
    if demo==True:
        price={}
        price["a"] = 24
        print(price)
        
        with open("ts.json","w") as f:
            json.dump(price,f,indent=4)

    filedict = {}
    data_dir = "G:/d_disk/airway_datasets"
    filenames = os.listdir(os.path.join(data_dir,"train_data_nii"))

    training_contents,validation_contents = [], []
    val_length = 0
    for filename in filenames:
        if val_length<(len(filenames)*0.2): # 如果验证集比例小于0.2
            if random.randint(1,2) == 1: # 并且随机数为1
                # 则放到验证集
                dic_content = {"image":f"train_data_nii/{filename}",
                                "label":f"train_label_nii/{filename}"}
                validation_contents.append(dic_content)
            else:
                # 否则放到训练集
                dic_content = {"image":f"train_data_nii/{filename}",
                                "label":f"train_label_nii/{filename}"}
                training_contents.append(dic_content)

        else:
            # 否则放到训练集
            dic_content = {"image":f"train_data_nii/{filename}",
                                "label":f"train_label_nii/{filename}"}
            training_contents.append(dic_content)
    
    
    labels_contents = {"0":"background","1":"airway"}
    filedict["labels"] = labels_contents
    
    filedict["training"] = training_contents
    filedict["validation"] = validation_contents

    

    with open("dataset.json","w") as f:
            json.dump(filedict,f,indent=4)

完成！