Inserting 3D objects into existing photographs
This fascinating video presents a new method to insert 3D objects into existing photographs. It is based on the research of Kevin Karsch, Varsha Hedau, David Forsyth and Derek Hoiem (all University of Illinois at Urbana-Champaign). Their main contribution is the algorithm, which generates the light model for the scene. The algorithm needs only one photograph and a few manual markings by a novice user together with a ground truth data set to create a near real life insertion. The ground truth data set was generated with 200 images from 20 indoor scenes under varying lighting conditions.
The video is well done and I am surprised whats possible, but I like to see how much user input is really necessary and how well the algorithm and the ground truth perform with other images. What do you think?
More details can be found at Kevin Karsch’s website.