CODEWITHSUNDEEP
pythonpytorchvectorizationtensor
Atul Balaji
Atul Balaji

Reputation: 812

Vectorized way to shuffle a given tensor using pytorch

I have a tensor A of shape (1,12,2,2) as follows: 

  ([[[[1., 3.],
      [9., 11.],

     [[ 2.,  4.],
      [10., 12.]],

     [[ 5.,  7.],
      [13., 15.]],

     [[ 6.,  8.],
      [14., 16.]],

     [[17., 19.],
      [25., 27.]],

     [[18., 20.],
      [26., 28.]],

     [[21., 23.],
      [29., 31.]],

     [[22., 24.],
      [30., 32.]],

     [[33., 35.],
      [41., 43.]],

     [[34., 36.],
      [42., 44.]],

     [[37., 39.],
      [45., 47.]],

     [[38., 40.],
      [46., 48.]]]])

I want to shuffle it using pytorch to produce the following tensor B of shape (1,3,4,4):

tensor([[[[ 1.,  6.,  3.,  8.],
          [21., 34., 23., 36.],
          [ 9., 14., 11., 16.],
          [29., 42., 31., 44.]],

         [[ 2., 17.,  4., 19.],
          [22., 37., 24., 39.],
          [10., 25., 12., 27.],
          [30., 45., 32., 47.]],

         [[ 5., 18.,  7., 20.],
          [33., 38., 35., 40.],
          [13., 26., 15., 28.],
          [41., 46., 43., 48.]]]])

I have implemented this using two for loops as follows: 

B = torch.zeros(1,3,4,4, dtype=torch.float)
ctr = 0
for i in range(2):
    for j in range(2):
        B[:,:,i:4:2,j:4:2] = A[:,ctr:ctr+3,:,:]
        ctr = ctr+3

I'm looking for any way to implement this in a vectorized manner in pytorch without these for loops. Maybe using functions like .permute(), etc.

Upvotes: 1

Views: 561

Answers (2)

Mohit Lamba
Mohit Lamba

Reputation: 1383

Just generalising the above solution for any upsampling factor 'r' like in pixel shuffle.

B = A.reshape(-1,r,3,s,s).permute(2,3,0,4,1).reshape(1,3,rs,rs)

Here 's' is the spatial resolution of each channel in 'A' and 'r' is the upsampling factor. For the particular case, r=2 and s=2. This solution should work for arbitary values of 'r' with appropriate size of 'A'.

So for the problem in hand s=2, r=2 and so the solution goes as

B = A.reshape(-1,2,3,2,2).permute(2,3,0,4,1).reshape(1,3,4,4)

as posted by @ddoGas

Similarly if 'A' had been of size (1, 192, 356, 532) and wanting to upsample by r=8 do

B = A.reshape(-1,8,3,356,532).permute(2,3,0,4,1).reshape(1,3,2848,4256)

Upvotes: 1

ddoGas
ddoGas

Reputation: 871

This will do the trick

B = A.reshape(2,2,3,2,2).permute(2,3,0,4,1).reshape(1,3,4,4)

Upvotes: 4

Related Questions