Chapter 29. Fast Graph Cuts for Computer Vision
P.J. Narayanan, Vibhav Vineet and Timo Stich
In this chapter we present an implementation of Graph Cuts with CUDA C. The computation pattern of this application is iterative with varying workloads and data dependencies between neighbors. We explore techniques to scale workload per iteration and how to parallelize the computation with the nontrivial data dependencies efficiently. This study shows the practical use of these techniques and reports the achieved performance on datasets for binary image segmentation and gives an outlook on how to extend the implementation to solve multilabel problems.
29.1. Introduction, Problem Statement, and Context
The work is done in the context of image segmentation. ...