计算pytorch标准化(Normalize)所需要数据集的均值和方差实例_Python

计算pytorch标准化(Normalize)所需要数据集的均值和方差实例

2020-04-21 09:37菜鸟川 Python

今天小编就为大家分享一篇计算pytorch标准化(Normalize)所需要数据集的均值和方差实例，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧

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：

				?

									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：

				?

									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))