Deep visual generation for automotive design upgrading and market optimising

Huang, Jingming (2023) Deep visual generation for automotive design upgrading and market optimising. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2023HuangJingMinPhD.pdf] PDF
Download (21MB)


The rising levels of homogeneity of modern cars in terms of price and functions has made exterior styling increasingly vital for market success. Recently, researchers have attempted to apply deep learning, especially deep generative models, to automotive exterior design, which has enabled machines to deliver diverse novel designs from large-scale data. In this thesis, we argue that recent advancements in deep learning techniques, particularly in deep generation, can be utilised to facilitate different aspects of automotive exterior design, including design generation, evaluation, and market profit predicting. We conducted three independent studies, each providing tailored solutions to specific automotive design scenarios. These include: a study focused on adapting the latest deep generative model to achieve regional modifications in existing designs, and evaluating these adjustments in terms of design aesthetic and prospective profit changes; another study dedicated to developing a predictive model to assess the modernity of existing designs regarding the future fashion trends; and a final study aiming to incorporate the distinctive shape characteristics of a cheetah into the side view designs of cars. This thesis has four main contributions. First, the developed DVM-CAR dataset is the first large-scale automotive dataset containing designs and marketing data over 10 years. It can be used for different types of research needs from multiple disciplines. Second, given the inherent constraints in automotive design, such as the need to maintain “family face”, and the fact that unconstrained design generation can be seen as a special form of regional modification, our research distinctively focuses on the regional modifications to existing designs, a departure from existing studies. Third, our studies are the first works that integrate the design modules with market profit optimisation. This reforms the traditional product design optimisation frameworks by replacing the abridged design profiles with graphical designs. Finally, the proposed data-driven measures offer effective approaches for automotive aesthetic evaluation and market forecasting, including approaches that can make assessments from a dynamic perspective by examining the evolving fashion trends.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Colleges/Schools: College of Science and Engineering > School of Computing Science
Supervisor's Name: Chen, Dr. Bowei
Date of Award: 2023
Depositing User: Theses Team
Unique ID: glathesis:2023-83877
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 13 Oct 2023 12:19
Last Modified: 13 Oct 2023 12:22
Thesis DOI: 10.5525/gla.thesis.83877
Related URLs:

Actions (login required)

View Item View Item


Downloads per month over past year