抄録
This study focuses on analyzing virtual images generated by AI to understand the characteristics of regional streetscapes, with a particular emphasis on visual information. Using Stable Diffusion, varied images of streetscapes with tree-lined avenues and buildings are generated, focusing on Pontocho, Shinsekai, and Midosuji. Evaluation involves impression assessment through surveys to gauge the streetscape authenticity perceived by participants, and image assessment through segmentation to determine the proportions of elements within images. Analyzing these results reveals how streetscape elements contribute to perceived authenticity. Multiple focal points by participants and the influence of material and condition variations per element were identified.
https://ror.org/01hvx5h04 