retrieveTestData.m

run 'configure';

pool=1000;
dir_name = './retrieved/';
mkdir(dir_name);
file_name = sprintf('./genPop1_5000/pop%d.mat', pool);
load(file_name);
base_pop = pop;
querys = {'beach';
'bird';
'boat';
'bridge';
'building';
'castle';
'cat';
'chair';
'dog';
'dress';
'fish';
'floor';
'flower';
'girl';
'grass';
'house';
'kitchen';
'mountain';
'office';
'river';
'road';
'rock';
'room';
'sand';
'sky';
'table';
'tower';
'tree';
'wall';
'water'};
for kk=1:size(querys, 1) 
    pop = base_pop;
    query = querys{kk,1};

    pool_name = sprintf([dir_name '/%d/' query '/'], pool);
    mkdir(pool_name);
    %% Load Dic
    % load([dic_path './dic.mat']);


    %% Load preprocessed test data
    disp('Loading data');
    % pair_name = sprintf('./testpair1.mat');
    % load([pair_path pair_name]);
    % bundle_name = sprintf('./testbundle1.mat');
    % load([bundle_path bundle_name]);
    % bundle = test_bundle;
    % clear test_bundle;
    % pair = test_pair;
    % clear test_pair;



    selectedHE = [];

    %% select hyperedges which has a quey in population
    split_query = regexp(query, '\s', 'split');

    if length(split_query) == 1
        count = 1;
        for i = 1:size(pop, 1)
            if sum(ismember(split_query, pop(i, 1).text)) >= 1
                selectedHE = [selectedHE;pop(i,1)];
            end
        end
    else
        count = 1;
        for i = 1:size(pop, 1)
            if sum(ismember(split_query, pop(i, 1).text)) >= length(split_query)
                buffer{count, 1} = pop(i, 1);
                count = count + 1;
                if count == BUFF_SIZE + 1
                    selectedHE = [selectedHE; buffer];
                    count = 1;
                end
            end
        end
    end

    pop = selectedHE;

    bundle_score = zeros(size(bundle, 1),1);    %similarity scores

    if isempty(selectedHE)
        disp([split_query ' is not exist']);
    else
        for m=1:i_order
            pop_bundle = zeros(size(pop, 1), K);
            pop_size = zeros(size(pop, 1), 1);            
            for i=1:size(pop,1)        
                if ~isempty(fieldnames(pop(i, 1).bundle)) 
                    pop_bundle(i, pop(i, 1).bundle(1,m).features) = 1;
                    pop_size(i, 1) = pop_size(i, 1)+length(pop(i, 1).bundle(1,m).features);
                    %pop_size(i, 1) = length(pop(i, 1).bundle.features);
                end
            end     
            pop_bundle = pop_bundle .* repmat(pop_size, 1, K);
            norm = sum(pop_bundle, 2);         
            for i=1:size(bundle, 1)
                img_bundle = zeros(K, size(bundle{i, 1}, 2)); 
                if size(fieldnames(bundle{i,1}), 1) < 2
                    continue;
                end
                for j = 1:size(bundle{i, 1}, 2)
                    img_bundle(bundle{i, 1}(j).features, j) = 1;
                end
                cmp_bundle = pop_bundle * img_bundle;
                cmp_bundle = cmp_bundle ./ repmat(norm, 1, size(bundle{i, 1}, 2));
                cmp_bundle(cmp_bundle < FEATURE_THRESHOLD) = 0;
                for j = 1:size(cmp_bundle, 1)
                    for k=1:size(cmp_bundle, 2)
                        hist_score = sum(sum(abs(pop(j, 1).bundle(m).hist - bundle{i, 1}(k).hist)));
                        hist_weight = sum(sum(pop(j, 1).bundle(m).hist, 2) ~= 0) / 255;     
                        cmp_bundle(j, k) = pop(i, 1).weight*(FEATURE_COEFF * cmp_bundle(j, k) + HIST_COEFF * hist_weight * (1 - hist_score));
                    end
                end              
                bundle_score(i,1) = bundle_score(i,1) + sum(sum(cmp_bundle));            
            end        
        end
    end
    [B, idx] = sort(bundle_score, 'descend');
    for i=1:20
        name = sprintf([pool_name '/%d_%d.jpg'], i, idx(i));
        imwrite(pair{idx(i),1}, name);
    end
end