PIL's Image or torch's read_image?

[gokulnotre5]

This doubt is in reference to loading a custom image (Custom Datasets). In the notebook, torch's read_image was used. I was wondering can't we use PIL's Image. Particularly,
from PIL import Image
image=Image.open("data/04-pizza-dan_dad.jpeg")
f_transform=simple_transform(image)
f=f_transform.reshape((1,3,64,64)).to(device)
class_names[model_0(f).argmax(dim=1).item()]

My question is, which one is better?