cnn_plate_setup_data.m

function imdb =cnn_plate_setup_data(datadir)
inputSize =[20,20,1];
subdir=dir(datadir);
imdb.images.data=[];
imdb.images.labels=[];
imdb.images.set = [] ;
imdb.meta.sets = {'train', 'val', 'test'} ;
image_counter=0;
trainratio=0.8;
for i=3:length(subdir)
        imgfiles=dir(fullfile(datadir,subdir(i).name));
        imgpercategory_count=length(imgfiles)-2;
        disp([i-2 imgpercategory_count]);
        image_counter=image_counter+imgpercategory_count;
        for j=3:length(imgfiles)
            img=imread(fullfile(datadir,subdir(i).name,imgfiles(j).name));
            img=imresize(img, inputSize(1:2));
            img=single(img);
%             [~,~,d]=size(img);
%             if d==3
%                 img=rgb2gray(img);
%                 continue;
%             end
            imdb.images.data(:,:,:,end+1)=single(img);
            imdb.images.labels(end+1)= i-2;
            if j-2<imgpercategory_count*trainratio
                imdb.images.set(end+1)=1;
            else
                imdb.images.set(end+1)=3;
            end
        end
end
dataMean=mean(imdb.images.data,4);
imdb.images.data = single(bsxfun(@minus,imdb.images.data, dataMean)) ;
imdb.images.data_mean = dataMean;
end