Pytorch圖像數據轉為Tensor張量

PyTorch的所有模型（nn.Module）都只接受Tensor格式的輸入，所以我們在使用圖像數據集時，必須將圖像轉換為Tensor格式。PyTorch提供了torchvision.transforms模塊來處理圖像數據集。torchvision.transforms模塊提供了一些常用的圖像預處理方法，如Resize、CenterCrop、RandomCrop、RandomHorizontalFlip等。
torchvision.transforms模塊還提供了ToTensor()方法，可以將PIL格式的圖像轉換為Tensor格式。ToTensor()方法會將圖像的像素值從[0, 255]范圍縮放到[0, 1]范圍，并將圖像的通道順序從(H, W, C)轉換為(C, H, W)。
本文以Pytorch中TorchVision的FashionMNIST數據集為例，展示如何將圖像數據轉換為Tensor格式。

import torch
from torchvision.datasets import FashionMNIST
from torchvision.transforms import ToTensor #將圖像數據轉換為張量 #加載數據集
train_data = FashionMNIST(root='./fashion_data', train=True, download=True) 
test_data = FashionMNIST(root='./fashion_data', train=False, download=True)

未轉換為Tensor格式的FashionMNIST數據集

train_data  #Dataset對象（輸入數據的集合） 60000個樣本

Dataset FashionMNISTNumber of datapoints: 60000Root location: ./fashion_dataSplit: Train

import matplotlib.pyplot as plt train_data = FashionMNIST(root='./fashion_data', train=True, download=True) img,clzz = train_data[5]  #返回一個元組，第一個元素是圖片，第二個元素是標簽
plt.imshow(img, cmap='gray') #顯示圖片,cmap='gray'表示以灰度圖顯示;img是一個tensor,是一個PIL.Image對象(python原始數據類型)
plt.title(clzz)
plt.show()

在這里插入圖片描述

train_data[1]  #返回的是一個元組，第一個元素是圖片，第二個元素是標簽
#train_data[1][0].shape  #圖像數據(一個顏色通道：28*28，圖像高度，圖像寬度)
#train_data[1][0].reshape(-1).shape  #將圖像數據展平為一維張量

(<PIL.Image.Image image mode=L size=28x28>, 0)

轉換為Tensor格式的FashionMNIST數據集

#加載數據集
train_data = FashionMNIST(root='./fashion_data', train=True, download=True, transform=ToTensor()) 
test_data = FashionMNIST(root='./fashion_data', train=False, download=True, transform=ToTensor())

train_data  #Dataset對象（輸入數據的集合） 60000個樣本

Dataset FashionMNISTNumber of datapoints: 60000Root location: ./fashion_dataSplit: TrainStandardTransform
Transform: ToTensor()

可以對比下圖像數據集轉換前后的效果，未轉換為Tensor格式的FashionMNIST數據集是PIL格式的圖像，而轉換為Tensor格式的FashionMNIST數據集是Tensor格式的圖像。

train_data[1]  #輸出的是一個元組，第一個元素是圖片，第二個元素是標簽

(tensor([[[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.0000, 0.0000, 0.1608, 0.7373, 0.4039, 0.2118, 0.1882, 0.1686,0.3412, 0.6588, 0.5216, 0.0627, 0.0000, 0.0000, 0.0000, 0.0000,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000, 0.0000, 0.1922,0.5333, 0.8588, 0.8471, 0.8941, 0.9255, 1.0000, 1.0000, 1.0000,1.0000, 0.8510, 0.8431, 0.9961, 0.9059, 0.6275, 0.1765, 0.0000,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0549, 0.6902, 0.8706,0.8784, 0.8314, 0.7961, 0.7765, 0.7686, 0.7843, 0.8431, 0.8000,0.7922, 0.7882, 0.7882, 0.7882, 0.8196, 0.8549, 0.8784, 0.6431,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.7373, 0.8588, 0.7843,0.7765, 0.7922, 0.7765, 0.7804, 0.7804, 0.7882, 0.7686, 0.7765,0.7765, 0.7843, 0.7843, 0.7843, 0.7843, 0.7882, 0.7843, 0.8824,0.1608, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.2000, 0.8588, 0.7804, 0.7961,0.7961, 0.8314, 0.9333, 0.9725, 0.9804, 0.9608, 0.9765, 0.9647,0.9686, 0.9882, 0.9725, 0.9216, 0.8118, 0.7961, 0.7961, 0.8706,0.5490, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.4549, 0.8863, 0.8078, 0.8000,0.8118, 0.8000, 0.3961, 0.2941, 0.1843, 0.2863, 0.1882, 0.1961,0.1765, 0.2000, 0.2471, 0.4431, 0.8706, 0.7922, 0.8078, 0.8627,0.8784, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.7843, 0.8706, 0.8196, 0.7961,0.8431, 0.7843, 0.0000, 0.2745, 0.3843, 0.0000, 0.4039, 0.2314,0.2667, 0.2784, 0.1922, 0.0000, 0.8588, 0.8078, 0.8392, 0.8235,0.9804, 0.1490, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.9686, 0.8549, 0.8314, 0.8235,0.8431, 0.8392, 0.0000, 0.9961, 0.9529, 0.5451, 1.0000, 0.6824,0.9843, 1.0000, 0.8039, 0.0000, 0.8431, 0.8510, 0.8392, 0.8157,0.8627, 0.3725, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.1765, 0.8863, 0.8392, 0.8392, 0.8431,0.8784, 0.8039, 0.0000, 0.1647, 0.1373, 0.2353, 0.0627, 0.0667,0.0471, 0.0510, 0.2745, 0.0000, 0.7412, 0.8471, 0.8314, 0.8078,0.8314, 0.6118, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.6431, 0.9216, 0.8392, 0.8275, 0.8627,0.8471, 0.7882, 0.2039, 0.2784, 0.3490, 0.3686, 0.3255, 0.3059,0.2745, 0.2980, 0.3608, 0.3412, 0.8078, 0.8118, 0.8706, 0.8353,0.8588, 0.8157, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.4157, 0.7333, 0.8745, 0.9294, 0.9725,0.8275, 0.7765, 0.9882, 0.9804, 0.9725, 0.9608, 0.9725, 0.9882,0.9922, 0.9804, 0.9882, 0.9373, 0.7882, 0.8314, 0.8824, 0.8431,0.7569, 0.4431, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0667, 0.2118, 0.6235,0.8706, 0.7569, 0.8157, 0.7529, 0.7725, 0.7843, 0.7843, 0.7843,0.7843, 0.7882, 0.7961, 0.7647, 0.8235, 0.6471, 0.0000, 0.0000,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.1843,0.8824, 0.7529, 0.8392, 0.7961, 0.8078, 0.8000, 0.8000, 0.8039,0.8078, 0.8000, 0.8314, 0.7725, 0.8549, 0.4196, 0.0000, 0.0000,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0235, 0.0000, 0.1804,0.8314, 0.7647, 0.8314, 0.7922, 0.8078, 0.8039, 0.8000, 0.8039,0.8078, 0.8000, 0.8314, 0.7843, 0.8549, 0.3569, 0.0000, 0.0118,0.0039, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0431,0.7725, 0.7804, 0.8039, 0.7922, 0.8039, 0.8078, 0.8000, 0.8039,0.8118, 0.8000, 0.8039, 0.8039, 0.8549, 0.3020, 0.0000, 0.0196,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0118, 0.0000, 0.0078,0.7490, 0.7765, 0.7882, 0.8039, 0.8078, 0.8039, 0.8039, 0.8078,0.8196, 0.8078, 0.7804, 0.8196, 0.8588, 0.2902, 0.0000, 0.0196,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0078, 0.0000, 0.0000,0.7373, 0.7725, 0.7843, 0.8118, 0.8118, 0.8000, 0.8118, 0.8118,0.8235, 0.8157, 0.7765, 0.8118, 0.8667, 0.2824, 0.0000, 0.0157,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0078, 0.0000, 0.0000,0.8431, 0.7765, 0.7961, 0.8078, 0.8157, 0.8039, 0.8118, 0.8118,0.8235, 0.8157, 0.7843, 0.7922, 0.8706, 0.2941, 0.0000, 0.0157,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.8314, 0.7765, 0.8196, 0.8078, 0.8196, 0.8078, 0.8157, 0.8118,0.8275, 0.8078, 0.8039, 0.7765, 0.8667, 0.3137, 0.0000, 0.0118,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.8000, 0.7882, 0.8039, 0.8157, 0.8118, 0.8039, 0.8275, 0.8039,0.8235, 0.8235, 0.8196, 0.7647, 0.8667, 0.3765, 0.0000, 0.0118,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.7922, 0.7882, 0.8039, 0.8196, 0.8118, 0.8039, 0.8353, 0.8078,0.8235, 0.8196, 0.8235, 0.7608, 0.8510, 0.4118, 0.0000, 0.0078,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.8000, 0.8000, 0.8039, 0.8157, 0.8118, 0.8039, 0.8431, 0.8118,0.8235, 0.8157, 0.8275, 0.7569, 0.8353, 0.4510, 0.0000, 0.0078,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,0.8000, 0.8118, 0.8118, 0.8157, 0.8078, 0.8078, 0.8431, 0.8235,0.8235, 0.8118, 0.8314, 0.7647, 0.8235, 0.4627, 0.0000, 0.0078,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.7765, 0.8157, 0.8157, 0.8157, 0.8000, 0.8118, 0.8314, 0.8314,0.8235, 0.8118, 0.8275, 0.7686, 0.8118, 0.4745, 0.0000, 0.0039,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.7765, 0.8235, 0.8118, 0.8157, 0.8078, 0.8196, 0.8353, 0.8314,0.8275, 0.8118, 0.8235, 0.7725, 0.8118, 0.4863, 0.0000, 0.0039,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,0.6745, 0.8235, 0.7961, 0.7882, 0.7804, 0.8000, 0.8118, 0.8039,0.8000, 0.7882, 0.8039, 0.7725, 0.8078, 0.4980, 0.0000, 0.0000,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000,0.7373, 0.8667, 0.8392, 0.9176, 0.9255, 0.9333, 0.9569, 0.9569,0.9569, 0.9412, 0.9529, 0.8392, 0.8784, 0.6353, 0.0000, 0.0078,0.0000, 0.0000, 0.0000, 0.0000],[0.0000, 0.0000, 0.0000, 0.0000, 0.0000, 0.0039, 0.0000, 0.0000,0.5451, 0.5725, 0.5098, 0.5294, 0.5294, 0.5373, 0.4902, 0.4863,0.4902, 0.4745, 0.4667, 0.4471, 0.5098, 0.2980, 0.0000, 0.0000,0.0000, 0.0000, 0.0000, 0.0000]]]),0)

此時，圖像數據集的每個圖像都是一個Tensor對象，Tensor對象的形狀為(C, H, W)，其中C表示圖像的通道數，H表示圖像的高度，W表示圖像的寬度。對于灰度圖像，C=1；對于RGB圖像，C=3。Tensor對象的dtype為torch.float32。
Tensor對象的像素值范圍為[0, 1]，而PIL格式的圖像的像素值范圍為[0, 255]。Tensor對象的像素值是浮點數，而PIL格式的圖像的像素值是整數。

train_data[1][0].shape  #圖像數據(一個顏色通道：28*28，圖像高度，圖像寬度)

torch.Size([1, 28, 28])

上面輸出的是FashionMNIST數據集中第0個圖像的Tensor對象。Tensor對象的形狀為(1, 28, 28)，表示圖像的通道數為1，高度為28，寬度為28。Tensor對象的dtype為torch.float32，表示數據類型為32位浮點數。Tensor對象的像素值范圍為[0, 1]，表示像素值是浮點數。

train_data[1][0].reshape(-1).shape  #將圖像數據展平為一維張量

torch.Size([784])