How-PIL-to-Tensor

How is the picture from $NumPy$ Form orientation $Tensor$ What about form conversion ?

from PIL import Image
import numpy as np
from torchvision.transforms import ToTensor
img=Image.open('./sword.png')
img_arr=np.array(img)
img_tensor=ToTensor().__call__(img)

$NumPy$ Picture format for $H,W,C$, namely $height,width,channels$( high 、 wide 、 The channel number )

img_arr.shape

(73, 183, 4)

$Tensor$ The format of the picture is $C,H,W$, namely $channels,height,width$( The channel number 、 high 、 wide )

img_tensor.shape

torch.Size([4, 73, 183])

Actually $NumPy$ Divide the corresponding pixel value at each position of the array by $255$ You get the corresponding $Tensor$ data type

#  Take out the first point in the upper left corner of the image RGBA value 
print(img_arr[0, 0, :])
print(img_tensor[:, 0, 0])

[ 35  33  47 255]
tensor([0.1373, 0.1294, 0.1843, 1.0000])

#  verification ,  Absolutely right 
print(255*img_tensor[:, 0, 0])
print(img_arr[0, 0, :])

tensor([ 35.,  33.,  47., 255.])
[ 35  33  47 255]