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)
