Home

Bastian Goldluecke

PLEASE NOTE

Documentation on the SourceForge web page is outdated and only valid up to and including release 5. Please visit the new project web page for the most recent information.




Download:



News (19.8.2014): Release 6 is out

See the new project web page for documentation and information.



News (11.6.2013): Release 5 is out

Release 5 is mostly devoted to the suite for variational light field analysis, but many of the core algorithms received fine tuning and bug fixes. A list of the major changes can be found below.

Build scripts have been extended to also work on Mac OS X and Fedora Core by Sergi Pujades, who also contributed improvements to the super-resolution codebase (thanks !).

Known issue: due to an oversight while compiling the download, the config files for the Middlebury data sets are incomplete, sorry - I'll update them as soon as I have time again (which, all things considered, might be in 25 years or so ;) ). However, the version in the svn trunk works.



Description:

COCOLIB is a library for variational image analysis, specifically continuous global optimization. It focuses on the minimization of functionals of the type

E(u) = J(u) + F(u)

where u is an image (i.e. vector-valued function on an interval), J is a convex and closed regularizer, and F a closed and convex data term.

The implementations were usually not designed for maximum speed, but to give a generic working reference implementation for the algorithms, which can be directly used in subsequent projects. However, for the most part, the code is reasonably fast.

All algorithms can be accessed from a command line tool with input parameters given as arguments or via configuration files, so no additional coding is necessary if you just want to try out the algorithms.

Since release 5, the project includes a fully functional command line tool for our light field analysis software, with example scripts to work on the HCI light field benchmark sets.



License:

cocolib is under GPL3 with an additional request: if you make use of the library or the command line tool in any form in a scientific publication, please refer to the cocolib web site and cite the paper

    @article{GSC12:siims,
      author = {B. Goldluecke and E. Strekalovskiy and D. Cremers},
      title = {The natural vectorial total variation which arises from geometric measure theory},
      journal = {SIAM Journal on Imaging Sciences},
      year = {2012},
    }



where the library is first mentioned officially. If you make use of the light field suite, please cite

    @article{WG13:tpami,
      author = {S. Wanner and B. Goldluecke},
      title = {Variational Light Field Analysis for Disparity Estimation and Super-Resolution },
      journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
      year = {2013},
    }


You should also cite the appropriate papers below when using the specific models, as per the list below.



The following are the major updates in release 5 (late June 2013):

  • Completely overhauled light field analysis suite to seamlessly work with new HCI light field benchmark data sets (automatic download scripts included). Newly supported algorithms include
    • Updated disparity estimation code with improvements in [14]
    • Updated super-resolution code with improvements in [14]
    • 4D light field and constrained disparity map denoising [15]
    • 4D light field inpainting [15]
    • 4D light field segmentation [16]

See our HCI web page for a description of the bench mark data sets.



The following models were added in release 4 (17.8.2012):

  • Multilabel regularizers [5,6]
    • Vectorial multi-label models with regularizers Potts, Linear (TV), Truncated Linear and Cyclic TV.
    • Segmentation data term and optical flow data term for vectorial multi-label problems
    • Test scripts to re-generate results for the journal version of [6], recently accepted for publication in SIIMS 2013.



The following models were added in release 3:

  • Multilabel regularizers

    • Total variation (linear) penalizer with arbitrary point-wise data term [8]
    • Relaxation of Potts penalizer with arbitrary point-wise data term [9,10]

  • Vectorial Total Variation regularizer:

    • Generic super-resolution data term [13]
    • Optical flow data term [11]



The following models were part of release 2:

  • Total Variation regularizer:

    • Linear data term (segmentation) [1]
    • Denoising data term (ROF model) [2,3]
    • Deblurring data term [1]
    • Inpainting data term (weighted ROF) [1]

  • Vectorial Total Variation regularizer [4]:

    • Denoising data term (Vectorial ROF and L^1)
    • Deblurring data term
    • Inpainting data term

  • Total Curvature regularizer [7]:

    • Denoising data term (TC-L2)
    • Deblurring data term
    • Inpainting data term
    • Segmentation data term (TC-Linear)



** Documentation **
Documentation is available in the project Wiki



Author

Project Admins:



contributions
Dierk Ole Johannsen - VTV optical flow
Sergi Pujades Rocamora - Mac OS X and Fedora Core support, improvements to the super-resolution codebase.



References:

Algorithm descriptions can be found in a variety of research papers, including

[1] Beck and Teboulle, "Fast gradient-based algorithms for constrained total variation image denoising and deblurring problems", IEEE Transactions on Image Processing 2009.

[2] Chambolle, "Total variation minimization and a class of binary MRF models", EMMCVPR 2005.

[3] Chambolle and Pock, "A first-order primal-dual algorithm for convex problems with applications to imaging", Mathematical Imaging and Vision, 2011.

[4] Goldluecke, Strekalovskiy and Cremers, "The natural vectorial total variation which arises from geometric measure theory". SIIMS 2012.

[5] Goldluecke and Cremers, "Convex Relaxation for Multilabel Problems with Product Label Spaces", ECCV 2010.

[6] Strekalovskiy, Goldluecke and Cremers, "Tight Convex Relaxations for Vector-Valued Labeling". ICCV 2011 and extended Technical Report, 2011 (available upon request).

[7] Goldluecke and Cremers, "Introducing Total Curvature for Image Processing", ICCV 2011.

[8] Pock, Cremers, Bischof, Chambolle, "Global Solutions of Variational Models with Convex Regularization", SIIMS 2010.

[9] Lellmann, Becker, Schnoerr, "Convex Optimization for Multi-Class Image Labeling with a Novel Family of Total Variation Based Regularizers", ICCV 2009

[10] Zach, Gallup, Frahm, Niethammer, "Fast Global Labeling for Real-Time Stereo Using Multiple Plane Sweeps", VMV 2008.

[11] Zach, Pock, Bischof. "A duality based approach for realtime TV-L1 optical flow", DAGM 2007.

[12] Wanner, Goldluecke, "Globally Consistent Depth Labeling of 4D Light Fields", CVPR 2012.

[13] Wanner, Goldluecke, "Spatial and Angular Variational Super-Resolution of 4D Light Fields", ECCV 2012.

[14] Wanner, Goldluecke, "Variational Light Field Analysis for Disparity Estimation and Super-Resolution", TPAMI 2013.

[15] Goldluecke, Wanner, "The Variational Structure of Disparity and Regularization of 4D Light Fields", CVPR 2013.

[16] Wanner, Straehle, Goldluecke, "Globally Consistent Multi-Label Assignment on the Ray Space of 4D Light Fields ", CVPR 2013.