For the reconstruction of 3D information from 2D images there are two principal approaches: manual or automatic.

Manual modeling CAD tools are often very powerful and flexible and they allow the user to produce models of very high quality. However the production costs are extremely high because the tools have to be operated by a well-trained designer.

Automatic algorithms on the other hand do not require a lot of human labor but they are often unable to automatically resolve ambiguities and they usually make assumptions about the quality of the input data which are not met in many practical scenarios.

Interactive reconstruction tools can emerge by replacing some steps of the manual reconstruction pipeline with an automatic algorithm or by modifying an automatic reconstruction scheme such that it can take user input into account in order to adjust parameters or set side constraints. A successful interactive reconstruction technique significantly accelerates the manual process while not compromising the output quality and it provides intuitive and realtime control on an automatic reconstruction scheme.

In my talk I will discuss this general approach based on a few example projects for image impainting, 3D object reconstruction as well as 3D city modeling where the overall production effort is minimized by splitting the problem into sub-tasks that are suitable for automation and those that can easily be controlled by the user.