Draft title: Mapping the Visual World Using Webcams, Cell Phones, and Satellites
Draft abstract: Trillions of pixels are collected every day. Together they supply many perspectives of the world, ranging from what someone had for dinner to where new buildings have been constructed. The key challenge is translating this raw data into useful information in a way that scales to large geographic areas. This talk provides an overview of our work in developing algorithms that learn to extract such information with limited human effort. Topics will include combining social media and satellite imagery to map visual attributes; inferring missing image metadata; and learning to synthesize novel views.