testfile.m

%% MULTIFRAME MOTION COUPLING FOR VIDEO SUPER RESOLUTION
% 
% 
% Be sure to initialize the submodules and to compile them
% following their instructions
%
%
% This test file shows how to use the MultiframeMotionCoupling class
% and how to change parameters, if needed.
%
%
% Please change the dataFolder and datasetName to the location of your dataset
% if you want to load a video as a folder of images.
% Alternatively you can remove that code and directly insert your video
% as a 4-D array in line 36.

clearvars;
addpath(genpath(cd))


%% Data properties
dataFolder = 'data/video_scenes';
datasetName = 'tube3';

startFrame = 1;
numFrames  = 3;

factor     = 4;             % Magnification factor

% Generate input data from ground truth data given
[imageSequenceSmall,imageSequenceLarge] = LoadImSequence([dataFolder,filesep,datasetName],startFrame,numFrames,factor,'bicubic');   



%% Construct algorithm object
% Input: RGB-Time matlab array 
mainSuper = MultiframeMotionCoupling(imageSequenceSmall);

%% Set variables (these are the standard parameters)

% Procedure
mainSuper.factor        = factor;              % magnification factor
mainSuper.verbose       = 1;                   % enable intermediate output, 1 is text, 2 is image
mainSuper.framework     = 'prost';             % Choose framework for super resolution problem
                                               % Either 'flexBox' or 'flexBox_vector'
                                               % or 'prost', if installed
mainSuper.comp_mode     = 'accurate';          % alternatives are 'fast' and 'fastest'
                                               % but only accurate directly replicates the results 
                                               % and timings from the paper
                                              

% Algorithm parameters (it is usually ok to keep these fixed for any videos)
mainSuper.alpha         = 0.01;                % regularizer weight
mainSuper.beta          = 0.2;                 % flow field complexity
mainSuper.kappa         = 0.25;                % regularization pendulum
mainSuper.flowDirection = 'backward';          % flow field direction
mainSuper.interpMethod  = 'average';           % Downsampling operator D
mainSuper.k = fspecial('gaussian',7,sqrt(0.6));% Blur operator B  

%% Init flow field and solvers
tic
mainSuper.init;
toc
%% Solve super resolution problem

tic
mainSuper.run;
toc
%% Show error margin

outImage = mainSuper.result1(20:end-20,20:end-20,:,ceil(numFrames/2));
psnrErr = round(psnr(outImage,imageSequenceLarge(20:end-20,20:end-20,:,ceil(numFrames/2))),2);
ssimErr = round(ssim(outImage,imageSequenceLarge(20:end-20,20:end-20,:,ceil(numFrames/2))),3);
disp(['PSNR (central patch, central slice): ',num2str(psnrErr),' dB']);
disp(['SSIM (central patch, central slice): ',num2str(ssimErr),' ']);


%% Visualize either central image or full video
if mainSuper.verbose > 0
     vid = implay(mainSuper.result1,2);  
     set(vid.Parent, 'Position',get(0, 'Screensize'));
else
    figure(1), imshow(outImage); title(['PSNR: ', num2str(psnrErr)]); axis image
end