Posted on Apr 27 • Edited on May 13

ToTensor in PyTorch

*Memos:

My post explains how to convert and scale a PIL image to an Image in PyTorch.
My post explains Compose().
My post explains ToImage().
My post explains ToDtype() about scale=True.
My post explains PILToTensor().
My post explains ToPILImage() about no arguments.
My post explains OxfordIIITPet().

ToTensor() can convert a PIL(Pillow library) image([H, W, C]), Image([..., C, H, W]) or ndarray to a tensor([C, H, W]) and scale the values of a PIL image or ndarray to [0.0, 1,0] as shown below:
*Memos:

ToTensor() is deprecated so instead use Compose(transforms=[ToImage(), ToDtype(dtype=torch.float32, scale=True)]) according to the doc.
A PIL image is scaled to [0.0, 1.0] if the PIL image belongs to one of the modes (L, LA, P, I, F, RGB, YCbCr, RGBA, CMYK, 1).
A ndarray is scaled to [0.0, 1.0] if the ndarray is uint8 which is [0, 255].
The 1st argument is img(Required-Type:PIL image, Image or tensor/ndarray(int/float/complex/bool)): *Memos:
- A tensor can be 0D or more D.
- A ndarray must be 2D or 3D.
- Don't use img=.
v2 is recommended to use according to V1 or V2? Which one should I use?.

from torchvision.datasets import OxfordIIITPet from torchvision.transforms.v2 import ToImage, ToTensor import torch import numpy as np ToTensor() # ToTensor()  PILImage_data = OxfordIIITPet( root="data", transform=None ) Image_data = OxfordIIITPet( root="data", transform=ToImage() ) Tensor_data = OxfordIIITPet( root="data", transform=ToTensor() ) Tensor_data # Dataset OxfordIIITPet # Number of datapoints: 3680 # Root location: data # StandardTransform # Transform: ToTensor()  Tensor_data[0] # (tensor([[[0.1451, 0.1373, 0.1412, ..., 0.9686, 0.9765, 0.9765], # [0.1373, 0.1373, 0.1451, ..., 0.9647, 0.9725, 0.9765], # ..., # [0.1098, 0.1098, 0.1059, ..., 0.2314, 0.2549, 0.2980]], # [[0.0784, 0.0706, 0.0745, ..., 0.9725, 0.9725, 0.9725], # [0.0706, 0.0706, 0.0784, ..., 0.9686, 0.9686, 0.9725], # ..., # [0.1059, 0.1059, 0.1059, ..., 0.3686, 0.4157, 0.4588]], # [[0.0471, 0.0392, 0.0431, ..., 0.9922, 0.9922, 0.9922], # [0.0392, 0.0392, 0.0471, ..., 0.9843, 0.9882, 0.9922], # ..., # [0.1373, 0.1373, 0.1373, ..., 0.8392, 0.9098, 0.8745]]]), 0)  Tensor_data[0][0].size() # torch.Size([3, 500, 394])  Tensor_data[0][0] # tensor([[[0.1451, 0.1373, 0.1412, ..., 0.9686, 0.9765, 0.9765], # [0.1373, 0.1373, 0.1451, ..., 0.9647, 0.9725, 0.9765], # ..., # [0.1098, 0.1098, 0.1059, ..., 0.2314, 0.2549, 0.2980]], # [[0.0784, 0.0706, 0.0745, ..., 0.9725, 0.9725, 0.9725], # [0.0706, 0.0706, 0.0784, ..., 0.9686, 0.9686, 0.9725], # ..., # [0.1059, 0.1059, 0.1059, ..., 0.3686, 0.4157, 0.4588]], # [[0.0471, 0.0392, 0.0431, ..., 0.9922, 0.9922, 0.9922], # [0.0392, 0.0392, 0.0471, ..., 0.9843, 0.9882, 0.9922], # ..., # [0.1373, 0.1373, 0.1373, ..., 0.8392, 0.9098, 0.8745]]])  Tensor_data[0][1] # 0  import matplotlib.pyplot as plt plt.imshow(X=Tensor_data[0][0]) # TypeError: Invalid shape (3, 500, 394) for image data  tt = ToTensor() tt(PILImage_data) # It's still PIL image. # Dataset OxfordIIITPet # Number of datapoints: 3680 # Root location: data  tt(PILImage_data[0]) # (tensor([[[0.1451, 0.1373, 0.1412, ..., 0.9686, 0.9765, 0.9765], # [0.1373, 0.1373, 0.1451, ..., 0.9647, 0.9725, 0.9765], # ..., # [0.1098, 0.1098, 0.1059, ..., 0.2314, 0.2549, 0.2980]], # [[0.0784, 0.0706, 0.0745, ..., 0.9725, 0.9725, 0.9725], # [0.0706, 0.0706, 0.0784, ..., 0.9686, 0.9686, 0.9725], # ..., # [0.1059, 0.1059, 0.1059, ..., 0.3686, 0.4157, 0.4588]], # [[0.0471, 0.0392, 0.0431, ..., 0.9922, 0.9922, 0.9922], # [0.0392, 0.0392, 0.0471, ..., 0.9843, 0.9882, 0.9922], # ..., # [0.1373, 0.1373, 0.1373, ..., 0.8392, 0.9098, 0.8745]]]), 0)  tt(PILImage_data[0][0]) # tensor([[[0.1451, 0.1373, 0.1412, ..., 0.9686, 0.9765, 0.9765], # [0.1373, 0.1373, 0.1451, ..., 0.9647, 0.9725, 0.9765], # ..., # [0.1098, 0.1098, 0.1059, ..., 0.2314, 0.2549, 0.2980]], # [[0.0784, 0.0706, 0.0745, ..., 0.9725, 0.9725, 0.9725], # [0.0706, 0.0706, 0.0784, ..., 0.9686, 0.9686, 0.9725], # ..., # [0.1059, 0.1059, 0.1059, ..., 0.3686, 0.4157, 0.4588]], # [[0.0471, 0.0392, 0.0431, ..., 0.9922, 0.9922, 0.9922], # [0.0392, 0.0392, 0.0471, ..., 0.9843, 0.9882, 0.9922], # ..., # [0.1373, 0.1373, 0.1373, ..., 0.8392, 0.9098, 0.8745]]])  tt(Image_data) # It's still `Image`. # Dataset OxfordIIITPet # Number of datapoints: 3680 # Root location: data # StandardTransform # Transform: ToImage()  tt(Image_data[0]) # (Image([[[37, 35, 36, ..., 247, 249, 249], # [35, 35, 37, ..., 246, 248, 249], # ..., # [28, 28, 27, ..., 59, 65, 76]], # [[20, 18, 19, ..., 248, 248, 248], # [18, 18, 20, ..., 247, 247, 248], # ..., # [27, 27, 27, ..., 94, 106, 117]], # [[12, 10, 11, ..., 253, 253, 253], # [10, 10, 12, ..., 251, 252, 253], # ..., # [35, 35, 35, ..., 214, 232, 223]]], dtype=torch.uint8,), 0)  tt(Image_data[0][0]) # Image([[[37, 35, 36, ..., 247, 249, 249], # [35, 35, 37, ..., 246, 248, 249], # ..., # [28, 28, 27, ..., 59, 65, 76]], # [[20, 18, 19, ..., 248, 248, 248], # [18, 18, 20, ..., 247, 247, 248], # ..., # [27, 27, 27, ..., 94, 106, 117]], # [[12, 10, 11, ..., 253, 253, 253], # [10, 10, 12, ..., 251, 252, 253], # ..., # [35, 35, 35, ..., 214, 232, 223]]], dtype=torch.uint8,)  plt.imshow(X=tt(PILImage_data[0][0])) plt.imshow(X=tt(Image_data[0][0])) # TypeError: Invalid shape (3, 500, 394) for image data  tt((torch.tensor(3), 0)) # int64 tt((torch.tensor(3, dtype=torch.int64), 0)) # (tensor(3), 0)  tt(torch.tensor(3)) # tensor(3)  tt((torch.tensor([0, 1, 2, 3]), 0)) # (tensor([0, 1, 2, 3]), 0)  tt(torch.tensor([0, 1, 2, 3])) # tensor([0, 1, 2, 3])  tt((torch.tensor([[0, 1, 2, 3]]), 0)) # (tensor([[0, 1, 2, 3]]), 0)  tt(torch.tensor([[0, 1, 2, 3]])) # tensor([[0, 1, 2, 3]])  tt((torch.tensor([[[0, 1, 2, 3]]]), 0)) # (tensor([[[0, 1, 2, 3]]]), 0)  tt(torch.tensor([[[0, 1, 2, 3]]])) # tensor([[[0, 1, 2, 3]]])  tt((torch.tensor([[[[0, 1, 2, 3]]]]), 0)) # (tensor([[[[0, 1, 2, 3]]]]), 0)  tt(torch.tensor([[[[0, 1, 2, 3]]]])) # tensor([[[[0, 1, 2, 3]]]])  tt((torch.tensor([[[[[0, 1, 2, 3]]]]]), 0)) # (tensor([[[[[0, 1, 2, 3]]]]]), 0)  tt(torch.tensor([[[[[0, 1, 2, 3]]]]])) # tensor([[[[[0, 1, 2, 3]]]]])  tt((torch.tensor([[0, 1, 2, 3]], dtype=torch.int32), 0)) # (tensor([[0, 1, 2, 3]], dtype=torch.int32), 0)  tt((torch.tensor([[0, 1, 2, 3]], dtype=torch.uint8), 0)) # (tensor([[0, 1, 2, 3]], dtype=torch.uint8), 0)  tt((torch.tensor([[0., 1., 2., 3.]]), 0)) # float32 tt((torch.tensor([[0., 1., 2., 3.]], dtype=torch.float32), 0)) # (tensor([[0., 1., 2., 3.]]), 0)  tt((torch.tensor([[0., 1., 2., 3.]], dtype=torch.float64), 0)) # (tensor([[0., 1., 2., 3.]], dtype=torch.float64), 0)  tt((torch.tensor([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]]), 0)) # complex64 tt((torch.tensor([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]], dtype=torch.complex64), 0)) # (tensor([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]]), 0)  tt((torch.tensor([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]], dtype=torch.complex32), 0)) # (tensor([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]], dtype=torch.complex32), 0)  tt((torch.tensor([[True, False, True, False]]), 0)) # bool tt((torch.tensor([[True, False, True, False]], dtype=torch.bool), 0)) # (tensor([[True, False, True, False]]), 0)  tt((np.array([[0, 1, 2, 3]]), 0)) # int32 tt((np.array([[0, 1, 2, 3]], dtype=np.int32), 0)) # (tensor([[[0, 1, 2, 3]]], dtype=torch.int32), 0)  tt((np.array([[[0, 1, 2, 3]]]), 0)) # int32 # (tensor([[[0]], [[1]], [[2]], [[3]]], dtype=torch.int32), 0)  tt((np.array([[0, 1, 2, 3]], dtype=np.int64), 0)) # (tensor([[[0, 1, 2, 3]]]), 0)  tt((np.array([[0, 1, 2, 3]], dtype=np.uint8), 0)) # (tensor([[[0.0000, 0.0039, 0.0078, 0.0118]]]), 0)  tt((np.array([[0, 1, 2, 3]], dtype=np.uint16), 0)) # (tensor([[[0, 1, 2, 3]]], dtype=torch.uint16), 0)  tt((np.array([[0., 1., 2., 3.]]), 0)) # float64 tt((np.array([[0., 1., 2., 3.]], dtype=np.float64), 0)) # (tensor([[[0., 1., 2., 3.]]], dtype=torch.float64), 0)  tt((np.array([[0., 1., 2., 3.]], dtype=np.float32), 0)) # (tensor([[[0., 1., 2., 3.]]]), 0)  tt((np.array([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]]), 0)) # complex128 tt((np.array([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]], dtype=np.complex128), 0)) # (tensor([[[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]]], dtype=torch.complex128), 0)  tt((np.array([[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]], dtype=np.complex64), 0)) # (tensor([[[0.+0.j, 1.+0.j, 2.+0.j, 3.+0.j]]]), 0)  tt((np.array([[True, False, True, False]]), 0)) # bool tt((np.array([[True, False, True, False]], dtype=bool), 0)) # (tensor([[[ True, False, True, False]]]), 0)

Top comments (0)

Some comments may only be visible to logged-in visitors. Sign in to view all comments. Some comments have been hidden by the post's author - find out more