How to lower the dimension of the tensor in pytorch?

1

I have tensor [1, 3, 256, 256, 3]. I want to reduce it to [1, 3, 256, 256]. How can i do this ?
Thanks!

2

How would you like to reduce the last dimension?
In a simple way you could just call x.mean(4) or another arithmetic operation.

1 Like
3

I could bring the tensor to the form [1, 3, 1, 256, 256], in numpy I would be able to reduce the dimension of np.squeeze and add another axis to the 0 position, but can I do it in pytorch?

4

Iā€™m not sure, which dimension you would like to squeeze or add, but PyTorch has also the method squeeze() and unsqueeze() to remove and add dimensions, respectively.

2 Likes
5

is there any way to reduce [1, 3, 256, 256, 3] to [1, 3, 256, 128, 3] ?

6

How would you like to reduce this dimension? Would you like to calculate the sum (or mean) of two neighboring values or just slice the tensor?

2 Likes
7

I solved it by using mean.
[1, 3, 256, 256, 3].view(1, 3, 2, 256*128, 3)
and then apply mean on third dimension.

1 Like
8

I have lowered torch dimension of shape torch.zeros([16, 3, 32, 32]) into [32,32,3] numpy array by

    img = image.squeeze(0).detach().cpu().numpy()
    print(img.shape) #(16, 3, 32, 32)
    img = img[0, :,:,:].transpose(1,2,0)
    print(img.shape # (32, 32, 3)
1 Like