The library cocolib implements a number of recent algorithms from continous convex optimization. In focus are linear inverse problems and multilabel problems for 2D images, as well as their counterparts for 4D light field analysis.
A detailed list of available methods follows, also note the reference page below.
Solvers for Linear Inverse Problems
- Generic algorithm implementations for FISTA  and Chambolle-Pock 
- Choice of data term (either L^1 or L^2 norm where applicable):
- denoising [2,3]
- deblurring 
- inpainting 
- super-resolution 
- Arbitrary choice of regularization:
- Vectorial total variation variants VTV_F, VTV_J, VTV_S (include normal TV as special case) 
- Total generalized variation TGV_2 , own color variant using VTV_F (probably reinvented)
- Total curvature 
- Coarse-to-fine scheme for optical flow 
Solvers for Continuous Multilabel Problems
- Total variation (linear) penalizer with arbitrary point-wise data term 
- Relaxation of Potts penalizer with arbitrary point-wise data term [9,10]
- Multiview stereo data term as a test case 
- Vectorial multi-label models with regularizers Potts, Linear (TV), Truncated Linear and Cyclic TV .
- Segmentation (PCMS) data term and optical flow data term for vectorial multi-label problems 
- Test scripts to re-generate results for 
Light field suite
- Disparity estimation code with improvements in 
- Super-resolution code with improvements in 
- 4D light field and constrained disparity map denoising 
- 4D light field inpainting 
- 4D light field segmentation 
- Seamless integration with benchmark datasets (HCI, Stanford, Middlebury) with automated download scripts
Command line tools
All algorithms are accessible via a command line interface, with parameters set in config files. The command line tool comes with at least one example configuration for each algorithm, which demonstrates the various options.
Release 6 (scheduled for August 2014) will include an interface to Matlab, which allows to use a subset of the algorithms directly in the interpreter. Currently, all inverse problems are supported with all variants of total variation regularizers implemented in cocolib (VTV_x and TGV_2). Furthermore, the interface allows rapid prototyping of new methods: it is possible to implement a data term directly in Matlab and pass it to the optimization algorithms in cocolib with arbitrary regularization.