Predicting and understanding Urban perception with convolutional neural networks

Cities' visual appearance plays a central role in shaping hu-man perception and response to the surrounding urban en-vironment. For example, the visual qualities of urban spaces affect the psychological states of their inhabitants and can induce negative social outcomes. Hence, it becomes criti-cally important to understand people's perceptions and eval-uations of urban spaces. Previous works have demonstrated that algorithms can be used to predict high level attributes of urban scenes (e.g. safety, attractiveness, uniqueness), ac-curately emulating human perception. In this paper we pro-pose a novel approach for predicting the perceived safety of a scene from Google Street View Images. Opposite to previous works, we formulate the problem of learning to predict high level judgments as a ranking task and we em-ploy a Convolutional Neural Network (CNN), significantly improving the accuracy of predictions over previous meth-ods. Interestingly, the proposed CNN architecture relies on a nov...