How to split a Pytorch tensor into different dimensions?

Question

I'm new to Pytorch .

Let's say I have a tensor that has this shape torch.size([1, 25200, 11])

I want to split it into 3 smaller tensors , each of 3 smaller tensors has the shape of 1st. torch.size([1, 3, 80, 80, 11]) and 2nd torch.size([1, 3, 40, 40 , 11]) and 3rd torch.size([1, 3, 20, 20, 11)].

Really appreciate your help.

Thanks

Explain those numbers:

80x80x3 = 19200

40x40x3 = 4800

20x20x3=1200 , add these result we have 25200, 1 is batch size, 11 is classes + xywh

Answer 1

Something like this should work.

import torch
tensor = torch.ones((1, 25200, 11))
first_break = tensor[:, 0:19200, :].view((1, 3, 80, 80, 11))
second_break = tensor[:, 19200:19200+4800, :].view((1, 3, 40, 40, 11))
third_break = tensor[:, 19200+4800:19200+4800+1200, :].view((1, 3, 20, 20, 11))

If you give a bit more explanation and context the code could get cleaned up and not be so hardcoded, or maybe this gives you enough to run with.

Answer 2

Have you tried:

  T1= torch.narrow(YourTensor, 1,0 , 80*80*3)
  T1v = T1.view(1,3,80,80,11)

  T2= torch.narrow(YourTensor, 1,80*80*3 , 40*40*3)
  T2v = T2.view(1,3,40,40,11)

  T3= torch.narrow(YourTensor, 1,80*80*3 + 40*40*3 , 20*20*3)
  T3v = T3.view(1,3,20,20,11)