Intelligent 3D modelling using evolution principle

As technologies advance over the years, machine learning techniques have advanced and are applied in different fields of application. Machine learning techniques are used to help perform the more intricate human task in a shorter amount of time. However, the time required to create a single new thre...

Full description

Saved in:
Bibliographic Details
Main Author: Ong, Wayne Chan Chi
Other Authors: Zheng Jianmin
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2020
Subjects:
Online Access:https://hdl.handle.net/10356/137989
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:As technologies advance over the years, machine learning techniques have advanced and are applied in different fields of application. Machine learning techniques are used to help perform the more intricate human task in a shorter amount of time. However, the time required to create a single new three-dimensional (3D) model that is creative is not proportional to the time and effort required by humans to perform the task. The purpose of this research is to investigate the effectiveness of using machine learning techniques in generating a new set of 3D models that are creative and of a wide variety within a shorter time frame. This is achieved by applying machine learning techniques to learn the feature representation of the components of a 3D model. Using BézierGAN and 3D VoxelGAN, the feature representation of the main body components and decoration components, respectively, from the 3D model is to be learned. The experiment analysed the effectiveness of the low dimensional latent representation in representing the data feed into the model during the training phase. Deformation will be performed on the component to ensure that components can be combined for the generation of new models. Results from the experiments show that the respective BézierGAN and 3D VoxelGAN have achieved the desired output after training was performed on the model. The feature representation of the individual data is captured in the latent representation obtained from the trained model. Therefore, this research definitively answers the question concerning the effectiveness of the application of machine learning techniques to generate new 3D models that are creative in a shorter period.