Editorial introduction to the CVIU special issue on 'Generative models in computer vision and medical imaging'

摘要

Generative models constitute a long-established paradigm for tackling computer vision problems, as their ability to represent complex objects and phenomena probabilistically, allows for a principled, Bayesian approach to vision, as well as facilitating easy visualization and evaluation by allowing us to reconstruct and synthesize the objects being modeled. Despite their broad success, generative models still pose great computational challenges, in particular when it comes to inferring their multiple hidden variables, and to learning the model parameters. Successfully tackling these problems can pave the way for the broader adoption of generative models in the computer vision and medical imaging community.