Using Gibbs Sampling to Estimate the Solution of the Unpaired Learning Problem

The article describes unpaired learning using Monte Carlo Markov Chain on the example of a stereo vision problem. The description includes the inference of the algorithm, the application of the stochastic gradient method, and some implementation details. Multiple penalty functions are considered, and quantitative results are presented. The results of the experiments expose new insights into weights for graphical models for stereo vision problems.