Access z-axis from tensor 4d

1

Hello everyone!
How can i get values from z-axis of a tensor 4d (example 32,128,4,4).
In this case i would like to get 128 value (planes) from the first 4x4 matrix and so on.
Anyone known the right operations to do that?

Thanks in advance!

2

I’m not sure I understand the use case correctly.
To access values in a tensor, you can use indexing as:

x = torch.randn(32, 128, 4, 4)
print(x[0, 0])

This would print the “first” 4x4 tensor.

Could you give an example, how your output should look like, please?

1 Like
3

Ok, i try to explain myself correctly.
In the image below, i would like to extract a tensor from each N plane (red points) starting from the first column/rown and so on. So, in this way i have to obtain X tensors composed by each planes (conv2). Is it possible with pytorch?

LeNet-5-1998
LeNet-5-19981552×595 41.2 KB

4

This is a sample:

import torch

x = torch.tensor(
    [
        [
            [[1, 2], [3, 4], ],
            [[5, 6], [7, 8], ],
            [[9, 10], [11, 12], ],
        ],
        [
            [[13, 14], [15, 16], ],
            [[17, 18], [19, 20], ],
            [[21, 22], [23, 24], ],
        ]
    ]
)

print(x.shape)
# torch.Size([2, 3, 2, 2])

print(x.reshape(2, 3, 4).transpose(2, 1).reshape(8, 3))
# tensor([[ 1,  5,  9],
#         [ 2,  6, 10],
#        [ 3,  7, 11],
#        [ 4,  8, 12],
#       [13, 17, 21],
#      [14, 18, 22],
#     [15, 19, 23],
#    [16, 20, 24]])

generally:

a, b, c, d = x.shape
x.reshape(a, b, c*d).transpose(2, 1).reshape(a*c*d, b)
1 Like
5

perfect, thanks a lot!

1 Like