ndimfun.m

Like cellfun and pagefun, but indifferent to which.

ndimfun

Like cellfun and pagefun, but indifferent to which. If input is a cell, use cellfun; if input is multi-page array, try pagefun / catch for-loop.

Sometimes cellfun, pagefun, and for-loops can be used interchangeably granted the data is formatted correctly for the use-case. I got annoyed having to write new loops every time I decided one format was better than the other, so I just smashed it all together into one function. Good for testing optimization of your code based on which version of the function you use.

Example: Image processing

Say you want to use @rot90 on a set of 20 intensity images of size 10 x 10.
You then may have a matrix of said imageset with size [10 10 20] or a cell of length 20 with arrays of size [10 10]
You might be wondering if it's computationally worth it to convert your imageset to a gpuArray and use pagefun or to just use cellfun
Use ndimfun to compare the processing times

Sample code:

dirs = fullfile({dir([fileparts(which('kobi.png')), '\AT3*.tif']).folder}, {dir([fileparts(which('kobi.png')), '\AT3*.tif']).name});
A = cellfun(@(x) im2gray(imread(x)), dirs, 'uniformoutput', false);
profile clear
profile on
a = ndimfun(@rot90, A);
aa = profile('info');
profile off
profile clear
profile on
sz = size(A{1});
B = gpuArray(reshape(cell2mat(A),sz(1), sz(2), []));
b = ndimfun(@rot90, B);
bb = profile('info');
profile off

disp(['cellfun time: ' num2str(sum([aa.FunctionTable.TotalTime])) ' s'])
disp(['gpuArray time: ' num2str(sum([bb.FunctionTable.TotalTime])) ' s'])