统计数据集图片通道的像素均值和方差

opencv读取图像为b,g,r方法,比如

img = cv2.imread(“xx.jpg”)
cv2.imshow(“xx”,img)

import os
import glob
import numpy as np
import cv2
import tqdm

def img_squeeze(img_input):
    img_out = img_input.reshape(1, -1)
    img_out = np.squeeze(img_out, 0)
    return img_out

filepath = r'./qiege/'
pathDirs = glob.glob(filepath + '/*.jpg')
pbar = tqdm.tqdm(total=len(pathDirs))
img_b = []
img_g = []
img_r = []
i=0
for pathDir in pathDirs:
    i += 1
    pathDir = pathDirs[int(np.random.rand(1)*(len(pathDirs)-1))]
    if i > 100:
        break
    pbar.update(1)
    img = cv2.imread(pathDir)
    img_b = np.concatenate((img_b, img_squeeze(img[:, :, 0])))
    img_g = np.concatenate((img_g, img_squeeze(img[:, :, 1])))
    img_r = np.concatenate((img_r, img_squeeze(img[:, :, 2])))
print(np.mean(img_b), np.mean(img_g), np.mean(img_r))
print(np.std(img_b), np.std(img_g), np.std(img_r))

尽量别统计太多，可能会爆内存，可以随机统计一些。
最终得到的结果分别是b,g,r
若要归一化，还要除以255。