Generalized Range Moves

TL;DR

This paper introduces a new move-making algorithm for energy minimization in multi-label Markov Random Fields that optimizes over all labels and variables simultaneously, leading to significant performance improvements.

Contribution

The proposed method enables simultaneous optimization over all labels and variables, surpassing the limitations of existing subset-based heuristics like alpha-expansion and range-moves.

Findings

Substantial improvement over previous algorithms.

Efficient optimization over all labels and variables.

Enhanced performance in energy minimization tasks.

Abstract

We consider move-making algorithms for energy minimization of multi-label Markov Random Fields (MRFs). Since this is not a tractable problem in general, a commonly used heuristic is to minimize over subsets of labels and variables in an iterative procedure. Such methods include {\alpha}-expansion, {\alpha}{\beta}-swap, and range-moves. In each iteration, a small subset of variables are active in the optimization, which diminishes their effectiveness, and increases the required number of iterations. In this paper, we present a method in which optimization can be carried out over all labels, and most, or all variables at once. Experiments show substantial improvement with respect to previous move-making algorithms.