463 / 2019-02-28 06:50:45

Exploring city-informative scenes and objects using street view images

How similar are cities look like? What are the most distinguished scenes and objects of a city? This work proposes a data-driven framework to learn and measure the visual knowledge of city appearance automatically by employing deep learning and computer vision. Based on the proposed framework, we compare the visual similarity and visual distinctiveness of 18 cities worldwide using more than two million social media photos. As a result, we identify the visual cues of each city that distinguish that city from others, such as landmarks, historical architecture, religious sites, unique urban scenes, and unusual natural landscapes; and explore a number of city-informative objects. Taking vehicles as an example, we find that the taxis, police cars and ambulances are the most city-informative objects. The results of this work are inspiring for various fields—providing insights on what large-scale geotagged data can achieve in place formalization and urban design.