makeData.m

load('boxes.mat');
load('synthBoxes.mat');
N = length(pictureDataFixed);
N_synth = length(synthPicsData);
binSize = 16;
M = binSize*binSize*binSize;
train_x = zeros(N, M);
train_t = zeros(N, 1);
synth_train_x = zeros(length(synthPicsData), M);
synth_train_t = zeros(length(synthPicsData), 1);

% for i=1:length(pictureDataFixed)
%     [hist_mat, hist_vec] = histogram(pictureDataFixed{i,1}, ...
%         pictureDataFixed{i,3}, binSize, @pixelNorm);
%     
%     % scaling to unit length
%     train_x(i,:) = hist_vec/sum(hist_vec);
%     train_t(i) = pictureDataFixed{i,2};
% end
% 
% % standardization
% train_x = bsxfun(@minus, train_x, mean(train_x));
% var = sum(train_x.^2)/(N-1);
% train_x = bsxfun(@rdivide, train_x, var.^0.5);
% train_x(isnan(train_x)) = 0;

for i=1:length(synthPicsData)
    [hist_mat, hist_vec] = histogram(synthPicsData{i,1}, ...
        synthPicsData{i,3}, binSize);
    
    % scaling to unit length
    synth_train_x(i,:) = hist_vec/sum(hist_vec);
    synth_train_t(i) = synthPicsData{i,2};
end

synth_train_x = bsxfun(@minus, synth_train_x, mean(synth_train_x));
var = sum(synth_train_x.^2)/(N_synth-1);
synth_train_x = bsxfun(@rdivide, synth_train_x, var.^0.5);
synth_train_x(isnan(synth_train_x)) = 0;


%save(sprintf('dataPixelsvm.mat'),'train_x','train_t','-v7.3');
save(sprintf('synthData.mat'),'synth_train_x','synth_train_t','-v7.3');