Learning to geolocate videos

摘要: A classifier training system trains classifiers for inferring the geographic locations of videos. number are provided, where each corresponds to a particular location and is trained from set videos that have been labeled as representing location. In one embodiment, further restricted those in which landmark matching label detected. The extracts, these videos, features characterize video, such audiovisual features, text address category features. Based on corresponding Each can be applied without associated labels predict whether, or how strongly, video represents prediction used variety purposes, automatic labeling with locations, presentation location-specific advertisements association display data relevant portions an electronic map.