| Title |
Analysis of Architectural Design Style Using CNN Output Layer Values |
| Authors |
이상현(Lee, Sang-Hyun) ; 한지후(Han, Ji-Hoo) |
| DOI |
https://doi.org/10.5659/JAIK.2023.39.3.23 |
| Keywords |
Parametric Design; Image classification; Architectural Style; Building Form; CNN; Frank Gehry; MVRDV |
| Abstract |
The purpose of this study is to present a methodology that can systematically evaluate whether there are morphological similarities commonly
found in the works of a specific architect. This work notes that the magnitude of the final output layer value of CNN applied to a particular
image implies the likelihood that the image can be classified into a particular category. To explore the morphological similarity or the
possibility of determining the existence of a style, the following process was performed. This was demonstrated through analysis of CNN
structures and empirical experiments that can evaluate the presence and strength of styles by the magnitude and deviation of the final output
values. A classifier model that distinguishes certain architect's works from those of other architects was created. The classifier model was
applied to the work of a specific architect to obtain the final output value for each work. The possibility of style evaluation using CNN by
comparing two architects who are often evaluated as strong in style and those who are not was confirmed. In this study, Frank Gehry, who
is evaluated as strong in style, and MVRDV, which is evaluated as weak in style were compared. In the case of Frank Gehry, it was
confirmed that the magnitude of the final output layer of the CNN model was larger and the deviation was smaller than those of the
MVRDV. Accordingly, it was proved that it is possible to evaluate the existence and strength of a style using the final output layer value of
the CNN model. |