The classification of architectural style is one of the most challenging problems in architectural history due to its temporal inter-class relationships between different styles and geographical variation within one style. Previous computer version approaches have primarily focused on general classification of multiple architectural styles based on historical age, but very few studies have attempted deep learning to address intra-class classification problems according to geographical location, which might reveal the significance of local evolution and adaption of ancient architectural style. Therefore, we exemplified gothic architecture as a certain genre and leased a new dataset containing gothic architecture in three different countries: France, England, and Italy. Besides, a trained model is susceptible to overfitting due to fecundity of regional parameters and shortcoming of dataset. In this paper, we propose a new approach to accurately classify intra-class variance in the sense of their geographical locations: visualization of Convolutional Neural Network. Experimentation on this dataset shows that the approach of intra-class classification based on local features achieves high classification rate. We also present interpretable explanations for the results, to illustrate architectural indication of intra-class classification.
CITATION STYLE
Wang, R., Gu, D., Wen, Z., Yang, K., Liu, S., & Jiang, F. (2019). Intra-class Classification of Architectural Styles Using Visualization of CNN. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11632 LNCS, pp. 205–216). Springer Verlag. https://doi.org/10.1007/978-3-030-24274-9_18
Mendeley helps you to discover research relevant for your work.