demo_testreal.m

function output = demo_testreal(iter,  blurimg_in, kernel,  pad)


caffe.reset_all();




model_path = 'models/';

solver_file = fullfile(model_path, 'llight_solver_test.prototxt');
save_file = fullfile(model_path, 'llight_ps256_bs1.mat');


Solver = modelconfig_test(solver_file, save_file, iter);


if ~exist('testnum', 'var')
   testnum = length(blur_data);
end



load ('GSVDgaudisk.mat');


pad_size = 55;

    
    if pad
        blurimg = padarray(blurimg_in,[pad_size, pad_size],'replicate','both');
    else
        blurimg = blurimg_in;
    end
    
    
    
    kernel = kernel / sum(kernel(:));
    kernelF = psf2otf(kernel, [151,151]);
    kernel = otf2psf(kernelF);
    IKernelF = conj(kernelF) ./ (conj(kernelF).*kernelF + 1/100);
    IKernel = otf2psf(IKernelF); 
    IKernel = L*L'*IKernel*R*R';    
    IKernel = IKernel / sum(IKernel(:));    %% estimated shrinked inverse kernel
    M = L'*IKernel*R;
    
    weights = Solver.Solver_.net.get_weights();
    for k = 1:50
        for j = 1:50
            weights(2).weights{1,1}(1,1,k,j) = M(k, j);
        end
    end
    Solver.Solver_.net.set_weights(weights);
    
    

    [row, col, cha] = size(blurimg);


    for cha = 1:3
        batchc = {single(blurimg(:,:,cha))};
        Solver.Solver_.net.blobs('data').reshape([row, col, 1, 1]);
        tic
        activec = Solver.Solver_.net.forward(batchc);
        toc
        num_output = length(activec);
        active= activec{num_output};
        output (:,:,cha) = active;
    end