cluster_resnet18.m

function [idx] = cluster_resnet18(img, xy, numclass, sposz, plotit)

%REF: https://doi.org/10.1101/2022.04.14.488259

if nargin < 4, plotit = false; end
if nargin < 3, numclass = 4; end
if nargin < 4, sposz = 40; end

try
    net = resnet18();
catch ME
    % disp(ME.message)
    answer = questdlg('This function requires the Deep Learning Toolbox Model for ResNet-18 Network support package for the pretrained weights. Install this support package?');
                if strcmp(answer, 'Yes')
                    matlab.addons.supportpackage.internal.explorer.showSupportPackages('RESNET18', 'tripwire');
                end
                rethrow(ME);
            end


            % ref: https://www.mathworks.com/matlabcentral/fileexchange/75342-image-clustering-and-dimension-reduction-using-cnn
            % net = resnet18();
            % net =resnet18('Weights','none');
            [imarray] = st_sampleimg(img, xy, sposz);
            fea = zeros(length(imarray), 1000);
            parfor kk = 1:length(imarray)
                fea(kk, :) = squeeze(activations(net, imresize(imarray{kk}, [224, 224]), 'fc1000'));
            end
            idx = kmeans(fea, numclass, "Start", "plus");

            if plotit
                figure;
                s = tsne(fea);
                scatter(s(:, 1), s(:, 2), [], idx);
                dt = datacursormode;
                dt.UpdateFcn = {@i_myupdatefcnx};
                fnm = tempname;
                save(fnm, 's', 'idx', 'imarray');
                fprintf('f=''%s.mat'';\ni_showimgsamples(f);\n', fnm);
            end

            function [txt] = i_myupdatefcnx(~, event_obj)
                persistent myupdatefcn3fig
                if isempty(myupdatefcn3fig) || ~isvalid(myupdatefcn3fig)
                    myupdatefcn3fig = figure('Position', [400, 400, 100, 100]);
                end
                if isvalid(myupdatefcn3fig) && isa(myupdatefcn3fig, 'matlab.ui.Figure')
                    figure(myupdatefcn3fig);
                end
                imagesc(imarray{event_obj.DataIndex});
                axis tight
                axis equal
                txt = {num2str(event_obj.DataIndex)};
            end

        end