2

I have a list of size (10,10,3,64)

This represents a list of 64 images of size 10x10x3

I want to create 1 image of size 80x80x3, where each image is side by side. I am not sure exactly how to merge these axes to make sure that the image integrity holds.

Just an example, a signle image is 

myList[all][all][all][0]
Mazdak
  • 105,000
  • 18
  • 159
  • 188
sookool99
  • 262
  • 2
  • 13

2 Answers2

3

IIUC one way to solve it would be with reshaping and permuting dimensions.

1) Reshape to split the last dimension into two dimensions.

2) Transpose to bring the last two split dims come next to the first two dims).

3) Finally reshape again to merge the first two dims and next two dims into one dim each.

Thus, we would have an implementation like so -

np.array(myList).reshape(10,10,3,8,8).transpose(0,3,1,4,2).reshape(80,80,3)
Divakar
  • 218,885
  • 19
  • 262
  • 358
1

You could try this through np.concatenate and nested list comprehensions, e.g.: 

arr = np.array(mylist)

arr2 = np.concatenate([ np.concatenate([arr[...,i + 8*j] for i in np.arange(8)],axis=0) for j in np.arange(8)],axis=1 )

print(arr2.shape)

gives

(80, 80, 3)

The list comprehensions split the full array of images into individual images stored in lists of length 8, and the concatenate then forms an array from these lists where each image is stored sequentially.

Note. It should be relatively easy to change the way you want to tile your images too by just changing the numbers used. For example if you wanted a 12x4 tiling instead of 8x8

arr2 = np.concatenate([ np.concatenate([arr[...,i + 12*j] for i in np.arange(12)],axis=0) for j in np.arange(4)],axis=1 )
Nick Walkden
  • 41
  • 2