一、voc数据集按比例划分train、val

下载的voc数据集 images ,annotations. 

经下代码 按比例划分成,只需要修改中文批注部分路

import os
import sys
import random
import shutil


if __name__ == '__main__':
    train_percent = 0.8  # 训练集比例
    val_percent = 0.2   # 验证集比例

    Root = '../datasets/voc-end/'  # 数据集文件夹路径
    imgs_path = Root + 'images/'    # 图片路径
    xml_path = Root + '/annotations/'  # 标注路径
    data_c = ['jpg', 'xml']     # 数据种类 如.jpeg需修改成jpeg
    
    imgs_list = []
    xml_list = []
    imgs_list = os.listdir(imgs_path)
    xml_list = os.listdir(xml_path)

    if len(imgs_list) != len(xml_list):
        sys.exit("图片数据集和标注不一致，请重新确认数据集！！！！")

    num = len(imgs_list)
    print(num)
    
    ftrain = open(Root+'train.txt', 'w')  # 训练集图片名称列表
    fval = open(Root+'val.txt', 'w')  # 验证集图片名称列表

    img_train_dir = imgs_path + "train/"
    img_val_dir = imgs_path + "val/"
    xml_train_dir = xml_path + "train/"
    xml_val_dir = xml_path + "val/"

    if not os.path.exists(img_train_dir):
        os.makedirs(img_train_dir)
    if not os.path.exists(img_val_dir):
        os.makedirs(img_val_dir)
    if not os.path.exists(xml_train_dir):
        os.makedirs(xml_train_dir)
    if not os.path.exists(xml_val_dir):
        os.makedirs(xml_val_dir)
    
    train_num = int(num*train_percent)
    train_sample = random.sample(imgs_list, train_num)
    # print(train_sample)
    # ------------------------------
    #     train img&xml move
    # ------------------------------
    for name in train_sample:
        shutil.move(imgs_path+name, img_train_dir+name)
        print(name)
        a, b = name.split('.')
        t_text = a + '\n'
        ftrain.write(t_text)
        shutil.move(xml_path+a+'.xml', xml_train_dir+a+'.xml')

    v_list = []
    for i in os.listdir(imgs_path):
        c = i[-3::]
        if c in data_c:
            v_list.append(i)
    # ------------------------------
    #     val img&xml move
    # ------------------------------
    for v_name in v_list:
        shutil.move(imgs_path+v_name, img_val_dir+v_name)
        print(v_name)
        a, b = v_name.split('.')
        v_text = a + '\n'
        fval.write(v_text)
        shutil.move(xml_path+a+'.xml', xml_val_dir+a+'.xml')
    # ------------------------------
    #       train & val list text
    # ------------------------------

    ftrain.close()
    fval.close()