I use the MNIST dataset to learn Pytorch.
This is from the documentation to get a picture.
import torch.nn.functional as F import torch from torchvision import datasets, transforms
Tensor comes from the torchvision dataset.
# Create prediction images, labels = next(iter(trainloader)) images[0].shape
This is the Tensor:
tensor([[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.2549, -0.5059, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -0.8667, -0.3961, 0.2471, 0.6784, 0.9922, 0.8824, 0.3882,
-0.8353, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.5922,
0.3490, 0.6235, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922,
-0.2235, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.3647, 0.9451,
0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.8510, 0.9922,
0.8745, -0.3725, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.0118, 0.9922,
0.9922, 0.9922, 0.9843, 0.8902, 0.3255, -0.1922, -0.6314, 0.7725,
0.9922, 0.7804, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 0.4431, 0.9922,
0.9922, 0.7725, -0.1608, -1.0000, -1.0000, -1.0000, -1.0000, 0.6235,
0.9922, 0.7804, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.0745, 0.9922,
0.7333, -0.7020, -1.0000, -1.0000, -1.0000, -1.0000, -0.3020, 0.9843,
0.9922, 0.7804, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.8980, 0.2078,
-0.0902, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.2863, 0.9922,
0.9922, 0.7804, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 0.0039, 0.9922,
0.9922, 0.7804, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -0.8275, 0.8275, 0.9922,
0.9922, -0.0118, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, 0.5765, 0.9922, 1.0000,
0.9529, -0.3725, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -0.9922, -0.3804, 0.9765, 0.9922, 0.9922,
0.0039, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -0.8667, 0.0275, 0.9922, 0.9922, 0.9922, 0.9922,
-0.4667, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
0.0039, 0.8980, 0.9137, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922,
0.9216, 0.8980, 0.8980, 0.8980, 0.8980, 0.8980, 0.6314, -0.1843,
-0.6941, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
0.6784, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922,
0.9922, 1.0000, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922,
0.8353, 0.1451, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
0.6784, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922,
0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.9922, 0.8745, 0.5059,
0.4196, -0.5686, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-0.1765, 0.9216, 0.9922, 0.9765, 0.6706, 0.6706, 0.2863, -0.4118,
-0.4118, -0.4118, -0.4118, -0.4118, -0.4118, -0.4118, -0.5529, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -0.4275, -0.2471, -0.2863, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000],
[-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000, -1.0000,
-1.0000, -1.0000, -1.0000, -1.0000]])
Out[181]: torch.Size([1, 28, 28])
I want to scale the image down to a 14×14 picture, so I guess I need a torch.Size([1, 14, 14])
I tried this, but it results in a different format:
F.interpolate(images[0], 14).shape Out[184]: torch.Size([1, 28, 14])
I expected this to work, but it results in an error:
F.interpolate(images[0], (14, 14))
ValueError: size shape must match input shape. Input is 1D, size is 2
Does anyone how to to get my desired result?
Advertisement
Answer
From the docs:
The input dimensions are interpreted in the form: mini-batch x channels x [optional depth] x [optional height] x width.
Currently the first 28 in your shape, and the 14 in the output size, are interpreted as the number of channels/colors, not the image height. Therefore the scaling does not happen in that dimension. Unsqueeze your input to be of shape (1, 1, 28, 28) to get the correct behavior.