数据集比例划分训练验证测试

说明：按比例划分数据集

适用于域自适应或域泛化根据不同于不同类别进行数据及自动划分

目录层级为：root/domains/class.

import os
import random
from tkinter import image_types
import shutil


trainval_percent = 0.9  # 训练和验证集所占比例，剩下的0.1就是测试集的比例
train_percent = 0.8  # 训练集所占比例，可自己进行调整
savePath_root = '.\\data'

root = 'E:\\data_lib\\Adaptiope\\modern_office31'
root_list = os.listdir(root)
for path in root_list:
    classes_path = os.path.join(root,path)
    class_list = os.listdir(classes_path)
    for Class in class_list:
        class_path = os.path.join(classes_path, Class)
        pic_files = os.listdir(class_path)

        num_train_val = int(len(pic_files) * trainval_percent)
        num_train = int(num_train_val * train_percent)

        train_val_sample = random.sample(pic_files, num_train_val)
        train_sample = random.sample(train_val_sample, num_train)

        for i in pic_files:
            if i in train_val_sample:
                if i in train_sample:
                    savePath_class = os.path.join(savePath_root, path, 'train', Class)
                    if not os.path.exists(savePath_class):
                        os.makedirs(savePath_class)
                    shutil.copy(os.path.join(class_path, i), savePath_class)
                else:
                    savePath_class = os.path.join(savePath_root, path, 'val', Class)
                    if not os.path.exists(savePath_class):
                        os.makedirs(savePath_class)
                    shutil.copy(os.path.join(class_path, i), savePath_class)
            else:
                savePath_class = os.path.join(savePath_root, path, 'test', Class)
                if not os.path.exists(savePath_class):
                    os.makedirs(savePath_class)
                shutil.copy(os.path.join(class_path, i), savePath_class)