Graphical models provide a ubiquitous modeling paradigm, which has been successfully employed up to now in a wide variety of computer vision tasks, including both low-level and high-level vision problems. Furthermore, due to their representational power, their modularity and (most importantly) due to their ability to efficiently capture dependencies or encode constraints, such models allow one to reason globally about an image or a scene. As such, they are expected to be of fundamental importance with regard to the task of natural scene understanding, which constitutes one of the central themes and goals of computer vision research. This workshop will bring together experts in the areas of graphical models and scene understading, where its two main goals will be to give a perspective of what has happened in the field lately, and to provide an overview of the next big challenges that need to be addressed (in terms of either modeling, inference or learning) that will help for making further progress towards natural scene understanding through an approrpiate use of graphical models.