PIL image나 np.array(PIL.Image)를 torch.tensor로 바꾸거나 vise versa 할 떄는 scaling을 조심해야함

가령 uint8 -> float32 would map the [0, 255] range into [0, 1] (and vice-versa).

PIL → Tensor

import torchvision.transforms as transforms
from PIL import Image

# load image in RGB mode (png files contains additional alpha channel)
img = Image.open('golf.png').convert('RGB')

# set up transformation to resize the image
resize = transforms.Resize([224, 224])
img = resize(img)
**to_tensor = transforms.ToTensor()**

# apply transformation and convert to Pytorch tensor
tensor = to_tensor(img)
# torch.Size([3, 224, 224])

# add another dimension at the front to get NCHW shape
tensor = tensor.unsqueeze(0)
# torch.Size([1, 3, 224, 224])

Conversions