计算pytorch标准化(Normalize)所需要数据集的均值和方差实例

时间:2022-06-18 23:06:54

pytorch做标准化利用transforms.Normalize(mean_vals, std_vals),其中常用数据集的均值方差有:

?
1
2
3
4
5
6
if 'coco' in args.dataset:
  mean_vals = [0.471, 0.448, 0.408]
  std_vals = [0.234, 0.239, 0.242]
elif 'imagenet' in args.dataset:
  mean_vals = [0.485, 0.456, 0.406]
  std_vals = [0.229, 0.224, 0.225]

计算自己数据集图像像素的均值方差:

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
import numpy as np
import cv2
import random
 
# calculate means and std
train_txt_path = './train_val_list.txt'
 
CNum = 10000   # 挑选多少图片进行计算
 
img_h, img_w = 32, 32
imgs = np.zeros([img_w, img_h, 3, 1])
means, stdevs = [], []
 
with open(train_txt_path, 'r') as f:
  lines = f.readlines()
  random.shuffle(lines)  # shuffle , 随机挑选图片
 
  for i in tqdm_notebook(range(CNum)):
    img_path = os.path.join('./train', lines[i].rstrip().split()[0])
 
    img = cv2.imread(img_path)
    img = cv2.resize(img, (img_h, img_w))
    img = img[:, :, :, np.newaxis]
    
    imgs = np.concatenate((imgs, img), axis=3)
#     print(i)
 
imgs = imgs.astype(np.float32)/255.
 
 
for i in tqdm_notebook(range(3)):
  pixels = imgs[:,:,i,:].ravel() # 拉成一行
  means.append(np.mean(pixels))
  stdevs.append(np.std(pixels))
 
# cv2 读取的图像格式为BGR,PIL/Skimage读取到的都是RGB不用转
means.reverse() # BGR --> RGB
stdevs.reverse()
 
print("normMean = {}".format(means))
print("normStd = {}".format(stdevs))
print('transforms.Normalize(normMean = {}, normStd = {})'.format(means, stdevs))

以上这篇计算pytorch标准化(Normalize)所需要数据集的均值和方差实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/weixin_38533896/article/details/85951903