用torchvision.transforms.normalize归一化图片

torchvision.transforms.normalize可以将图片像素值归一化到指定的均值和标准差。

示例代码：

import torch
import torchvision.transforms as transforms
from PIL import Image

# 加载图片并转换为Tensor
image = Image.open('example.jpg')
transform = transforms.Compose([
    transforms.ToTensor()
])
image = transform(image)

# 归一化图片
normalize = transforms.Normalize(
    mean=[0.485, 0.456, 0.406],  # ImageNet数据集的均值
    std=[0.229, 0.224, 0.225]   # ImageNet数据集的标准差
)
image = normalize(image)

# 输出归一化后的图片像素值
print(image)

输出结果：

tensor([[[ 0.1613,  0.1613,  0.1613,  ...,  0.9556,  0.9556,  0.9556],
         [ 0.1613,  0.1613,  0.1613,  ...,  0.9556,  0.9556,  0.9556],
         [ 0.1613,  0.1613,  0.1613,  ...,  0.9556,  0.9556,  0.9556],
         ...,
         [-1.2886, -1.2886, -1.2886,  ..., -1.0040, -1.0040, -1.0040],
         [-1.2886, -1.2886, -1.2886,  ..., -1.0040, -1.0040, -1.0040],
         [-1.2886, -1.2886, -1.2886,  ..., -1.0040, -1.0040, -1.0040]],

        [[ 0.1260,  0.1260,  0.1260,  ...,  0.8757,  0.8757,  0.8757],
         [ 0.1260,  0.1260,  0.1260,  ...,  0.8757,  0.8757,  0.8757],
         [ 0.1260,  0.1260,  0.1260,  ...,  0.8757,  0.8757,  0.8757],
         ...,
         [-1.1549, -1.1549, -1.1549,  ..., -0.8273, -0.8273, -0.8273],
         [-1.1549, -1.1549, -1.1549,  ..., -0.8273, -0.8273, -0.8273],
         [-1.1549, -1.1549, -1.1549,  ..., -0.8273, -0.8273, -0.8273]],

        [[-0.0225, -0.0225, -0.0225,  ...,  0.2446,  0.2446,  0.2446],
         [-0.0225, -0.0225, -0.0225,  ...,  0.2446,  0.2446,  0.2446],
         [-0.0225, -0.0225, -0.0225,  ...,  0.2446,  0.2446,  0.2446],
         ...,
         [-0.5113, -0.5113, -0.5113,  ..., -0.3837, -0.3837, -0.3837],
         [-0.5113, -0.5113, -0.5113,  ..., -0.3837, -0.3837, -0.3837],
         [-0.5113, -0.5113, -0.5113,  ..., -0.3837, -0.3837, -0.3837]]])