How to alternatively concatenate pytorch tensors?

user_15623388 0
Viewed 45

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

pytorch deep-learning
1 Answers

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]])
user_167331010
Related