Friday, September 08, 2006

3D Photo Tours

With all the recent web coverage of Microsoft Photosynth, I thought that I should post some of its foundational work here.

From the paper:

"Our system consists of an image-based modeling front end, which automatically computes the viewpoint of each photograph as well as a sparse 3D model of the scene and image to model correspondences. Our photo navigation tool uses image-based rendering techniques to smoothly transition between photographs, while also enabling full 3D navigation and exploration of the set of images and world geometry, along with auxiliary information such as overhead maps. We demonstrate our system on several large personal photo collections as well as images gathered from photo sharing Web sites on the Internet."
From a computer vision standpoint, they claim to be using a modified SIFT for wide-baseline matching. While this does not appear to be revolutionary results, it should be understood that, in this case, presentation is everything. This work appears to have a very intuitive interface that could revolutionize image search/retrieval/browsing. Check out this Interactive Demo of a trimmed-down version. Also, here is a nice video (120MB MOV) from SIGGRAPH.

Full Paper (PDF)