Richard Szeliski Microsoft Research 3D structure from motion and multi-view stereo techniques have made tremendous advances in the last decade and it is now possible to reconstruct city-wide 3D models from collections of photos found on the web. This talk reviews incremental and large-scale structure from motion, focusing both on technique that work reliably over a wide range of input images and acquisition conditions, as well as techniques that scale for extremely large datasets consisting of thousands of images.