Image-based Rendering and Modeling

Richard Szeliski
Microsoft Research

Image-based rendering and image-based modeling (aka 3D reconstruction from images) are two sides of the same coin. IBM strives to create accurate and photorealistic models from collections of images, whereas IBR tries to render detailed photorealistic scenes from these models along with the original imagery. These fields have made dramatic progress in the last two decades. Fully automated sparse and dense reconstruction techniques are now used in a variety of applications such as 3D aerial map building. IBR systems such as interactive panoramas and navigable street-level imagery are used by millions of people to explore the world. This talk surveys the history of the field, including fundamental representations and algorithms as well as open problems such as the modeling and rendering of reflections.

Presentation (PDF File)

Back to Graduate Summer School: Computer Vision