Finding "retro" places in Japan: crowd-sourced urban ambience estimation

Understanding the ambience of an area is essential for making various geographical decisions. This kind of ambience (e.g., "beautiful," "quiet," "happy," "retro," etc.) is due to not only physical features, such as scenery and functionality, but also subjective information based on people's perceptions and experiences. Although scholars in various fields have attempted to quantitatively measure subjective human experiences, the amount of location-based subjective data varies depending on the area. Thus, we have proposed a method to quantify individuals' perceptions of urban ambiences, as well as a means to anticipate these perceptions from landscape images. In this study, we have focused on the "retro" ambience of old Japanese cities. First, we used crowdsourcing to assign ambience scores to the cities of Nara and Kyoto, the ancient capitals of Japan, based on the level of perception of retro people in the landscape images. Further, we trained a deep learning model to estimate the ambience score of unknown landscape images using previously labelled data. Finally, we discussed the utility of our method for routing applications and exploring its generalizability across different cities.