pytorch torchvision.transforms 对图像数据进行归一化

原创

于 2020-03-23 09:09:32 发布 · 4.1k 阅读

7 ·

CC 4.0 BY-SA版权

本文详细介绍了如何使用PyTorch及其torchvision库对numpy数组和PIL图像进行归一化处理，展示了不同转换代码的效果，并解释了ToTensor()函数的工作原理。

1.对numpy数据进行归一化

import torch
import torchvision
import torchvision.transforms as transforms
import numpy as np

X = np.random.choice(255, (5,4,3))
print(X)

transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize(mean=(0,0,0), std=(1,1,1)),
    ])

transform(X)

print(X)
print(X.shape)

2.对PIL读取的图像进行归一化

import torch
from PIL import Image
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy

plt.rcParams['font.sans-serif']=['SimHei'] #用来正常显示中文标签


img = Image.open('D:\datasets\img\lena.jpg')
plt.figure()
plt.imshow(img)
plt.title('原图',fontsize=12,color='r')
plt.show()

transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize(mean=(0,0,0), std=(8,8,8)),
    ])

img1 = transform(img)
img1 = np.array(img1*255, dtype=np.uint8)

print('转换后的shape', img.shape)
img_np = np.zeros((200,200,3),dtype=np.uint8)