深層学習と進化的アルゴリズムを組み合わせた木漏れ日を感じるデザインの生成

抄録

This study proposes an optimization method based on qualitative scores, specifically focusing on the similarity to nature-like features. The key to the proposed method is the utilization of a deep neural network trained for classifying depth images of objects, including natural objects. An evolutionary algorithm is employed to explore the parameters of a generative algorithm to minimize the loss. A test themed on “Komorebi-like-design" demonstrated a strong correlation between the losses and the results of a paired comparison method experiment using images generated by the proposed method.