python-返回数据集路径，并计算数据集mean、std

# -*- coding: utf-8 -*-
import os
def all_files_path(rootDir):
    for root, dirs, files in os.walk(rootDir):     # 分别代表根目录、文件夹、文件
        for file in files:                         # 遍历文件
            file_path = os.path.join(root, file)   # 获取文件绝对路径
            filepaths.append(file_path)            # 将文件路径添加进列表
        # for dir in dirs:                           # 遍历目录下的子目录
        #     dir_path = os.path.join(root, dir)     # 获取子目录路径
        #     all_files_path(dir_path)               # 递归调用
if __name__ == "__main__":
    filepaths = []                                 # 初始化列表用来
    all_files_path('文件路径')
    with open('******.txt', 'a') as f:
        for filepath in filepaths:
            f.write(filepath + '\n')

import os
from PIL import Image
import numpy as np
import tqdm


def main():
    # 数据集通道数
    img_channels = 3
    # txt文件中每一行是一个图片路径
    txt = '***.txt'
    assert os.path.exists(txt), f"'{txt}' does not exist."
    # 遍历txt每一行获取路径
    imgs_path = []
    with open(txt, 'r') as f:
        for line in f:
            imgs_path.append(line.strip('\n'))
    # 累计mean和std，三个通道，这里是RGB，PIL库中的Image.open 默认RGB，cv2.imread是BGR
    cumulative_mean = np.zeros(img_channels)
    cumulative_std = np.zeros(img_channels)
    # 统计数据集长度
    print(f"INFO: {len(imgs_path)} imgs in total")
    for img_path in tqdm.tqdm(imgs_path, total=len(imgs_path)):
        img = np.array(Image.open(img_path)) / 255.
        # 对每个维度进行统计，Image.open打开的是HWC格式，最后一维是通道数
        for d in range(3):
            cumulative_mean[d] += img[:, :, d].mean()
            cumulative_std[d] += img[:, :, d].std()

    mean = cumulative_mean / len(imgs_path)
    std = cumulative_std / len(imgs_path)
    print(f"mean: {mean}")
    print(f"std: {std}")


if __name__ == '__main__':
    main()