【生成无标签数据集格式】

对于无监督任务

1，假设整个数据集文件夹如下：

data_root
├── meta
│ ├── test.txt # 测试数据集的标注文件
│ ├── train.txt # 训练数据集的标注文件
│ └── val.txt # 验证数据集的标注文件
├── train
│ ├── 123.png
│ ├── folder_1
│ │ ├── xxx.png
│ │ └── xxy.png
│ └── nsdf3.png
├── test
└── val

示例代码：

import os
import random
from sklearn.model_selection import train_test_split
import shutil

def gen_unlabel_dataset(image_folder, data_root):
    # 创建meta目录及其子目录
    meta_dir = os.path.join(data_root, "meta")
    os.makedirs(meta_dir, exist_ok=True)

    # 创建train、test和val目录
    train_dir = os.path.join(data_root, "train")
    test_dir = os.path.join(data_root, "test")
    val_dir = os.path.join(data_root, "val")

    os.makedirs(train_dir, exist_ok=True)
    os.makedirs(test_dir, exist_ok=True)
    os.makedirs(val_dir, exist_ok=True)
    
    # 获取图片文件列表
    Image_suffix = ['jpg', 'bmp', 'png', 'jpeg']
    image_files = []
    for img in os.listdir(image_folder):
        if img.split('.')[-1] in Image_suffix:
            image_files.append(img)
    # 随机打乱图片文件夹列表
    random.shuffle(image_files)
    
    # 划分训练集和验证集
    train_files, val_files = train_test_split(image_files, test_size=0.2)
    
    # 将文件夹写入train.txt
    train_file = os.path.join(meta_dir, "train.txt")
    with open(train_file, "w", encoding="utf-8") as f:
        for file in train_files:
            f.write(f"{file}\n")
            
            # 将图片复制到train目录下
            src_path = os.path.join(image_folder, file)
            dst_path = os.path.join(train_dir, file)
            shutil.copy(src_path, dst_path)

    # 将文件路径写入val.txt
    val_file = os.path.join(meta_dir, "val.txt")
    with open(val_file, "w", encoding="utf-8") as f:
        for file in val_files:
            f.write(f"{file}\n")
            # 将图片复制到val目录下
            src_path = os.path.join(image_folder, file)
            dst_path = os.path.join(val_dir, file)
            os.rename(src_path, dst_path)

    # 创建test.txt
    test_file = os.path.join(meta_dir, "test.txt")
    open(test_file, "w").close()

if __name__ == "__main__":
    image_folder = r''
    data_root = os.path.join(image_folder, "unlabeled_dataset")
    os.makedirs(data_root, exist_ok=True)