怎么计算数据的均值和方差

news2024/9/28 13:17:49

1. 前言

pytorch 跑深度学习数据集的时候，需要先将数据归一化，可以让网络更好的收敛。一般的均值和方差都是利用（灰度值 - 0.5 ）/ 0.5

本章提供一个计算均值和方差的代码，供以后使用

2. 完整代码

代码如下：

import numpy as np
import os
from PIL import Image
import cv2


# 输入 PyTorch的 dataset, 输出均值和标准差
def compute_mean_and_std(path):
    img_path = [os.path.join(path,i) for i in os.listdir(path)]         # 所有数据的绝对路径

    mean_r,mean_g,mean_b = 0,0,0

    for i in img_path:
        img = np.array(Image.open(i).convert('RGB'))
        img = cv2.cvtColor(img,cv2.COLOR_RGB2BGR)

        mean_b += np.mean(img[:, :, 0])               # 所有数据的灰度值相加
        mean_g += np.mean(img[:, :, 1])
        mean_r += np.mean(img[:, :, 2])

    mean_b /= len(path)             # 平均灰度
    mean_g /= len(path)
    mean_r /= len(path)

    N = 0
    diff_r, diff_g, diff_b = 0, 0, 0
    for i in img_path:
        img = np.array(Image.open(i).convert('RGB'))
        img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

        diff_b += np.sum(np.power(img[:, :, 0] - mean_b, 2))
        diff_g += np.sum(np.power(img[:, :, 1] - mean_g, 2))
        diff_r += np.sum(np.power(img[:, :, 2] - mean_r, 2))
        N += np.prod(img[:, :, 0].shape)

    std_b = np.sqrt(diff_b / N)
    std_g = np.sqrt(diff_g / N)
    std_r = np.sqrt(diff_r / N)

    mean = (mean_b.item() / 255.0, mean_g.item() / 255.0, mean_r.item() / 255.0)
    std = (std_b.item() / 255.0, std_g.item() / 255.0, std_r.item() / 255.0)
    return np.round(mean,3),np.round(std,3)


if __name__ == '__main__':
    root = './data'                 # 数据的根目录
    mean,std = compute_mean_and_std(path=root)
    print(mean)
    print(std)