A Multiscale Framework for Challenging Discrete Optimization
This addresses a specific bottleneck in discrete optimization for tasks like image processing or computer vision, but appears incremental as it builds on existing multiscale approaches.
The paper tackles the problem of challenging contrast-enhancing discrete energies in optimization, where current methods struggle, by proposing a multiscale framework with an energy-aware interpolation operator, resulting in significant improvement over state-of-the-art methods.
Current state-of-the-art discrete optimization methods struggle behind when it comes to challenging contrast-enhancing discrete energies (i.e., favoring different labels for neighboring variables). This work suggests a multiscale approach for these challenging problems. Deriving an algebraic representation allows us to coarsen any pair-wise energy using any interpolation in a principled algebraic manner. Furthermore, we propose an energy-aware interpolation operator that efficiently exposes the multiscale landscape of the energy yielding an effective coarse-to-fine optimization scheme. Results on challenging contrast-enhancing energies show significant improvement over state-of-the-art methods.