A generative A.I. technique for structure and texture transfer in furniture design

• Main Author: Gallega, Rgee Wharlo
• Format: text
• Language: English
• Published: Animo Repository 2021
• Subjects: Furniture design; Artificial intelligence; Neural networks (Computer science); Artificial Intelligence and Robotics; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etdm_comsci/2; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1004&context=etdm_comsci
• Institution: De La Salle University

Recently, generative AI methods such as style transfer and generative adversarial networks (GANs) have made significant advancements in the general area of design, where outputs (i.e. shoes, buildings, bags) are generated with near-realistic quality. Furthermore there are studies that have applied these methods into the area of furniture design. Generative models including GANs enable synthesizing new furniture and interpolation of structural features between furniture; on the other hand, style transfer allows transferring the textures and colors from one furniture to another. Based on style transfer, we propose a generative AI technique that performs both texture transfer and structure transfer from furniture images onto 3D furniture models. Our technique was evaluated on transferring textural and structural features from a dataset of Filipino designer furniture images onto chair and table models from the ShapeNet dataset. Experimental results showed that our texture transfer method outperformed all baselines in texture synthesis based on Wasserstein Distance. We recommend further investigation on improving image-to-model structure transfer, and also on ways for texture transfer to be more applicable in the industry. Keywords: Texture Transfer, Texture Synthesis, Structure Transfer, Style Transfer, Furniture Design, Generative AI, Convolutional Neural Networks, Deep Learning