How to turn 3D image matrix to 2d matrix without a for loop? Python and numpy

Question

Currently working on a lab from a Data course at school. What I want to do is convert matrix_a with shape (26,64,64) to a new matrix_b (64,1664). The (64,64) inside matrix_a is the bits that make up a series of images, and the (64,1664) matrix_b should result in a strip of the images. I tried using np.reshape, which does reshape the matrix correctly, but the images are lost due to the ordering used. I could use a for loop to iteratively insert each 64x64 image into matrix_b, yet they're asking that you do not use a for loop. They mentioned something about using splicing? I'm writing this in python with numpy. Apologies if this post makes no sense it's my first one. Thanks

What 26,64,64 matrix represents? 64x64 image on 26 different time steps or something different? You should provide some insights about the matrix representations. — Alperen Kantarcı, Aug 08 '20 at 17:34
Related:[numpy with python: convert 3d array to 2d](https://stackoverflow.com/questions/32838802/numpy-with-python-convert-3d-array-to-2d), — wwii, Aug 08 '20 at 18:24

score 1 · Answer 1 · answered Aug 08 '20 at 17:46

1

With numpy, you can try this (assume data_3d is your 3d array):

data_2d = data_3d.swapaxes(1,2).reshape(3,-1)

answered Aug 08 '20 at 17:46

ldren

159
5

hammi · Answer 2 · 2021-01-24T12:21:11.427

0

import numpy as np

arr = np.random.randint(10, size= [26,64,64])
arr.shape

>>> (26, 64, 64)

transpose() reorders axes from [0, 1, 2] to [1, 0, 2].

arr = arr.transpose([1,0,2])   # edit suggested by @wwii 
arr.shape

>>> (64, 64, 26)

arr = arr.reshape([64, -1])
arr.shape

>>> (64, 1664)

edited Jan 24 '21 at 12:21

answered Aug 08 '20 at 18:17

hammi

804
5
14

OP wants a 2d array with the 26 64x64 images *concatenated*. – wwii Aug 08 '20 at 18:20
@wwii "and the (64,1664) matrix_b should result in a strip of the images" exact words of OP – hammi Aug 08 '20 at 18:21
The shape might be correct but does the data come out correct? Try it with (4,3,3) array - `a = np.arange(3*3*4).reshape(4,3,3)`. – wwii Aug 08 '20 at 18:41
1

`arr..transpose([1,0,2])` looks better. – wwii Aug 08 '20 at 18:59
@wwii yeah the `transpose()` takes care of that. OP wasnt specific about how he wanted to strip data column wise or row wise, thus two options for `transpose()` – hammi Aug 08 '20 at 20:11

wwii · Accepted Answer · 2020-08-08T19:10:31.950

>>> a = np.arange(2*2*3).reshape(3,2,2)
>>> a
array([[[ 0,  1],
        [ 2,  3]],

       [[ 4,  5],
        [ 6,  7]],

       [[ 8,  9],
        [10, 11]]])

Concatenate each image.

Horizontal strip - which is what you asked for in the question.

>>> np.concatenate(list(a),-1)
array([[ 0,  1,  4,  5,  8,  9],
       [ 2,  3,  6,  7, 10, 11]])

Vertical strip

>>> np.concatenate(list(a),0)
array([[ 0,  1],
       [ 2,  3],
       [ 4,  5],
       [ 6,  7],
       [ 8,  9],
       [10, 11]])
>>>

np.hstack(tuple(a)) and np.vstack(tuple(a)) produce identical results to concatenate.

np.vsplit(a,a.shape[0]) is equivalent to list(a) or tuple(a).

How to turn 3D image matrix to 2d matrix without a for loop? Python and numpy

3 Answers3