Abstract
Thanks to the rapid development of simulation techniques and the popularization of social network services, it becomes increasingly convenient to utilize heterogeneous data produced and shared by non-expert users to solve problems in computer graph-ics and computer vision. This thesis focuses on exploring the feasibility of devising data-driven optimization approaches to achieve desirable design and perception re-sults about cities.To begin with, we discuss the innovations that data-driven optimization approaches have brought to the field of urban modeling and understanding, illustrated with the advantages of relevant applications. Then we describe the novelty of data-driven opti-mization by proposing solutions to three pertinent problems. Specifically, we apply data-driven optimization approaches to the design of mid-scale layouts and urban transport networks in the context of urban modeling from computer graphics perspec-tive. Regarding urban understanding from computer vision perspective, we utilize data-driven optimization approaches to generate urban zoning maps from non-VHR satellite imagery data and geo-tagged photos.