Table 5. Results for multilabel Potts datasets. The first column indicates the name of a dataset and the number of instances. What follows is a comparison between multiscale and single-scale inference of Lazy-Flipper, as well as a comparison to the “Ace” method for each dataset. We report the average energy (value) and run-time (time), the percentage of instances on which an algorithm reported the best energy (best) and provided a certificate of global optimality (enclosed in brackets). Note that nearly all of the instances are solved to optimality, and that different inference methods perform well on different datasets.
...read more