Slides: Photo Tourism (part 1), IM2GPS (part 2)

The Internet has become an increasingly massive, interesting, and useful source of imagery for computer vision and graphics applications. In this course, we will explore recent developments in 3D vision for Internet photo collections. We will focus on two systems, Photo Tourism and im2gps, which aim to automatically recover geometry and location information from these diverse, unstructured sets of photos. These systems help us answer such questions as: where was this photo taken? What was it looking at?

The first part of this course will describe Photo Tourism. Photo Tourism is a structure from motion (SfM) system for reconstructing camera positions and scene geometry from unorganized photo collections, and demonstrated that SfM techniques can be successfully applied to photos downloaded from the Internet. We will cover the basics of the structure from motion problem, and talk about the design and implementation considerations of Photo Tourism that are critical for handling unordered collections robustly and efficiently. Photo Tourism also enables new interfaces for exploring and visualizing photo collections in 3D. We will talk about how to take a set of reconstructed and cameras and points and create immersive 3D experiences.

Structure from motion generally recovers relative scene geometry; how can we figure out where in the world the photos were taken? The second part of the course will focus on the im2gps system for geolocating photographs. We will discuss the use of Internet image data sources and related logistical issues, the geolocation of single images or sequences of photos, and the use of geography estimates for deeper image understanding tasks such as object detection.

Course content:

About the organizers:
Noah Snavely is an Assistant Professor in the Department of Computer Science at Cornell University. Originally from Tucson, Arizona, he received a B.S. in Computer Science from the University of Arizona in 2003. He then worked with Steven M. Seitz and Richard Szeliski at the University of Washington, receiving his Ph.D. in 2008. Noah is interested in computer vision and computer graphics, especially in using vast amounts of imagery from the Internet for reconstructing and visualizing the world.

James Hays received his B.S. in Computer Science from Georgia Institute of Technology in 2003. He is a Ph.D. student in Carnegie Mellon University's Computer Science Department and is advised by Alexei A. Efros. His research interests are in computer vision and computer graphics, focusing on image understanding and manipulation leveraging massive amounts of data. His research has been supported by a National Science Foundation Graduate Research Fellowship. He will graduate this summer.