Pytorch - select region of a tensor using torch function

Question

I am looking for a way to select a region of a PyTorch tensor using a torch function (without using numpy). Do you have suggestions on how to proceed?

In other words, I'm looking for a way to crop a region of a matrix. Using numpy, it would be something like

import numpy as np
A = np.random.rand(16,16)
B = A[0:8, 0:8]

The approach I am trying is the following:

from torchvision import transforms
A = torch.randn([1,3,64,64])
B = torch.split(A, [16,32,16], dim =2)
C = torch.split(B, [16,32,16], dim =3)

Which gives the error

'tuple' object has no attribute 'split'

Shai · Accepted Answer · 2020-03-04 14:14:15Z

1

What's wrong with regular slicing?

import torch

A = torch.randn([1,3,64,64])
B = A[..., 16:32, 16:32]

answered Mar 4, 2020 at 14:14

Shai

115k39 gold badges259 silver badges398 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

albusdemens Over a year ago

For some weird reason I have to avoid that... I ended up using torch.nn.functional.pad, like A = torch.nn.functional.pad(D, (0,-32,0,-32)).

Shai Over a year ago

@albus_c this is indeed a twisted workaround. If you can explain what prevents you from using regular indices, we might find a better/simpler solution. pytorch is quite flexible (I think as flexible as numpy in this case).

Collectives™ on Stack Overflow

Pytorch - select region of a tensor using torch function

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related