How can I create a torch tensor from a numpy.array

Question

I have created a code that generates a matrix with shape (12,12) for color image analyzes using convolutional neural network. The input of my script is a torch tensor of shape (5,3,12,12). I took off the values 5 and 3 using detach().numpy().

The script :

    for k in range(5):
            for l in range(3):
                y=x[k][l].detach().numpy()
                m,n= y.shape
                im=np.pad(y,((1,1),(1,1)),'constant')
                Enhanced = np.zeros((m,n))
                for i in range(1,m+1):
                    for j in range(1,n+1):
                        ...      
                z.append(Enhanced)

... represents a simple function which I don't want to bother you with.

z is a list of the Enhanced which are numpy arrays. so my goal is to create a torch tensor from the Enhanced numpy arrays with shape (5,3,12,12).

I added this line to my code inside the for loop and i get:

 r=torch.Tensor(z)
 print(r.shape)

and then print the r.shape and i get that :

 torch.Size([3, 12, 12])
 torch.Size([6, 12, 12])
 torch.Size([9, 12, 12])
 torch.Size([12, 12, 12])
 torch.Size([15, 12, 12])

So what I understand is that I need to stack those r tensors. I used the function t=np.stack(r) but what I get is the shape of the last execution which is torch.Size([15, 12, 12]) so how can I modify that to get shape of (5, 3, 12, 12)

Szymon Maszke · Accepted Answer · 2021-07-13 13:26:19Z

3

You have to:

stack list of np.array together (Enhanced ones)
convert it to PyTorch tensors via torch.from_numpy function

For example:

import numpy as np

some_data = [np.random.randn(3, 12, 12) for _ in range(5)]

stacked = np.stack(some_data)
tensor = torch.from_numpy(stacked)

Please note that each np.array in the list has to be of the same shape

For different shapes one could do that:

import numpy as np
import torch

some_data = [np.random.randn(3, 12, 12) for _ in range(5)] + [
    np.random.randn(6, 12, 12)
]

stacked = np.concatenate(some_data).reshape(-1, 3, 12, 12)
tensor = torch.from_numpy(stacked)

print(tensor.shape)

In your case:

r = torch.from_numpy(np.concatenate(z).reshape(-1, 3, 12, 12))

edited Jul 13, 2021 at 13:26

answered Jul 13, 2021 at 12:48

Szymon Maszke

25.2k4 gold badges54 silver badges92 bronze badges

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

Collectives™ on Stack Overflow

How can I create a torch tensor from a numpy.array

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related