Understanding the visual discrepancy and heterogeneity of different places is of great interest to architectural design, urban design and tourism planning. However, previous studies have been limited by the lack of adequate data and efficient methods to quantify the visual aspects of a place. This work proposes a data-driven framework to explore the place-informative scenes and objects by employing deep convolutional neural network to learn and measure the visual knowledge of place appearance automatically from a massive dataset of photos and imagery. Based on the proposed framework, we compare the visual similarity and visual distinctiveness of 18 cities worldwide using millions of geo-tagged photos obtained from social media. As a result, we identify the visual cues of each city that distinguish that city from others: other than landmarks, a large number of historical architecture, religious sites, unique urban scenes, along with some unusual natural landscapes have been identified as the most place-informative elements. In terms of the city-informative objects, taking vehicles as an example, we find that the taxis, police cars and ambulances are the most place-informative objects. The results of this work are inspiring for various fields-providing insights on what large-scale geo-tagged data can achieve in understanding place formalization and urban design.