SelfSimSR

Exploiting Self-Similarities for Single Frame Super-Resolution (ACCV 2010)

Chih-Yuan Yang Email address: cyang35 [at] ucmerced [dot] edu 11/11/2010 v1.0 first release 02/23/2013 v1.1 The files are reorganized and the modified sample.cpp file is included. 09/22/2013 v1.2 The ann_sample.exe is rebuilt containing static linked MFC library. I encounter a problem that some machines do not have the libarary so that our MATLAB code fails on calling the external ann_sample.exe file. I also updated the Visual Studio 2010 project file sample.vcxproj in the folder Lib\ann_1.1.1\MS_Win32\sample. The rebuilt ann_sample.exe in the folder Lib\ann_1.1.1\MS_Win32\bin has been copied to the folder used by our MATLAB code: Lib\ann_1.1.1\ANN_Windows

================================================== How to start

The Test_SanityTest.m file works as a sanity test. The results are saved at the folder TempFolder\SanityTest with the file name SanityTest_Iter(X).png. The resolution of SanityTest_input.png is 30x30. On a machine with a CPU as Intel i7 2.67GHz, the algroithm takes 5 miniutes to generate the first output image. The computational load of the algorithm increases exponentially due to the resolution of the input image. For example, the resolution of Child_input.png is 128x128 pixels. To run the Test_Child.m file on a machine with a Intel i7 3.6GHz CPU, the algorithm needs 10 hours to generate the first output image, and additional 7 hours to generate the next image. The most time consuming step is the computation of the group sparse coefficients for many groups.

================================================== Tested platforms

This package is tested on two machines. Machine 1 OS: Windows 7 64 bits MATLAB version: MATLAB 2012b CPU: Intel i7 920 2.67GHz Memory: 24GB

Machine 2 OS: Ubuntu 12.04 64 bits MATLAB version: MATLAB 2012b CPU: Intel i7-3820 3.6GHz Memory: 32GB

================================================== About the scaling factor

v1.1 only supports scaling factors 3 and 4 as I hard-coded some variables.

================================================== About the libraries

1. Ann package The Ann library used in this package is v1.1.1 originally download from http://www.cs.umd.edu/~mount/ANN. I modified the original ann_sample.cpp to dump ann results to be loaded by the MATLAB. The modified ann_sample.cpp is in the folder ModifiedLibraryFiles

2. K-SVD package This package is written by M. Aharon, M. Elad, and A.M. Bruckstein, and there is a Readme.txt in the package folder.

3. spgl1-1.7 package The Spectral Projected Gradient for L1 minimization package contains a Readme in its own folder.