How to convert Matlab cell array into Python numpy array?

In matlab, I generate a random # of images of the shape (130,100) and save them one by one in a cell array. I then need to import this cell array of images into python and convert them into a numpy array (number_images, 130,100).

I've tried the following, but I get an error. Not sure how to fix. Would appreciate help- thanks!

import scipy.io as sio
import numpy
from PIL import Image

folder='insert path of mat file here'
imgs = sio.loadmat(folder+'/img_array.mat')
num_imgs=len(imgs['img_array'][0])
img_array=np.array(imgs['img_array'][0])
new_array=img_array.reshape((num_images,130,100))

ValueError: cannot reshape array of size 9 into shape (9,130,100)

**In this case, 9 images were generated from matlab, and this was the cell array shown.

Answer 1

[Al Danial]

One way to get the 3D array of images in Python is to convert the cell array to a 3D matrix on the matlab side and save that (instead of saving a cell array of 2D matrices). It would be nice if cell2mat() preserved the 3D structure but it doesn't so a bit of reshaping in matlab is needed. A fortuitous side effect is that matlab's column-major storage flips nicely into NumPy's row-major storage. Here's an example using a cell array of two images, each of which is 3x4:

MATLAB:

>> img_array = { reshape(1:12, 3,4) reshape(13:24, 3,4) };  % cell array of two matrices
>> img_array{:}

ans =
1     4     7    10
2     5     8    11
3     6     9    12

ans =
13    16    19    22
14    17    20    23
15    18    21    24

>> cell2mat(img_array)    % all matrices concatenated horizontally to a large 2D matrix

ans =
1     4     7    10    13    16    19    22
2     5     8    11    14    17    20    23
3     6     9    12    15    18    21    24

>> three_d = reshape(cell2mat(img_array), 4,3,2)  % gives the transpose of original images

three_d(:,:,1) =
1     5     9
2     6    10
3     7    11
4     8    12

three_d(:,:,2) =
13    17    21
14    18    22
15    19    23
16    20    24

save('cell_array.mat','three_d')

Then in Python the .mat file loads directly into the original 3D array:

Python:

In : from scipy.io import loadmat
In : x = loadmat('cell_array.mat')
In : x['three_d'].T
Out:
array([[[ 1,  2,  3,  4],
        [ 5,  6,  7,  8],
        [ 9, 10, 11, 12]],
    
       [[13, 14, 15, 16],
        [17, 18, 19, 20],
        [21, 22, 23, 24]]], dtype=uint8)