CODEWITHSUNDEEP
deep-learningpytorch
Nick Young
Nick Young

Reputation: 23

How to alternatively concatenate pytorch tensors?

Pytorch provides API to concatenate tensors, like cat, stack. But does it provide any API to concatenate pytorch tensors alternatively?

For example，enter image description here

suppose input1.shape = C*H*W, a1.shape = H\*W, and output.shape = (3C)*H*W

This can be achieved using a loop, but I am wondering if any Pytorch API can do this

Upvotes: 2

Views: 376

Answers (1)

Hamzah Al-Qadasi
Hamzah Al-Qadasi

Reputation: 9806

I will try to do it with small example:

input1 = torch.full((3, 3), 1)
input2 = torch.full((3, 3), 2)
input3 = torch.full((3, 3), 3)

out = torch.concat((input1,input2, input3)).T.flatten()
torch.stack(torch.split(out, 3), dim=1).reshape(3,-1)

#output

tensor([[1, 2, 3, 1, 2, 3, 1, 2, 3],
        [1, 2, 3, 1, 2, 3, 1, 2, 3],
        [1, 2, 3, 1, 2, 3, 1, 2, 3]])

Upvotes: 1

Related Questions