All of these applications require detailed and realistic models of a real-world scene's geometry and the way it reflects light. Creating such models is however a difficult and resource consuming task often requiring expensive equipment such as 3D scanning systems and a large amount of manual work.
In this talk, I will focus on image-based acquisition techniques that are an efficient way to capture and model real world scenes. I will first give a brief overview of my work on image-based appearance acquisition for a variety of material types including specular and translucent objects. I will then focus on the reconstruction of accurate geometry models from a set of images using a robust multi-view stereo approach. The system's robustness makes it applicable to very general datasets captured under uncontrolled conditions such as outdoor images taken with different cameras at different times and under varying weather conditions. I will demonstrate this by reconstructing the geometry of famous scenes using collections of tourist photos gathered from Internet imaging sites. I will furthermore show the accuracy of the reconstructed geometry models using standard benchmark datasets.