EPSRC logo

Details of Grant 

EPSRC Reference: EP/V057022/1
Title: Procedural Geometry for Humans
Principal Investigator: Kelly, Dr T
Other Investigators:
Researcher Co-Investigators:
Project Partners:
Department: Sch of Computing
Organisation: University of Leeds
Scheme: New Investigator Award
Starts: 01 November 2021 Ends: 31 October 2024 Value (£): 382,018
EPSRC Research Topic Classifications:
Artificial Intelligence Computer Graphics & Visual.
Fundamentals of Computing Information & Knowledge Mgmt
Software Engineering
EPSRC Industrial Sector Classifications:
Creative Industries
Related Grants:
Panel History:
Panel DatePanel NameOutcome
13 Sep 2021 EPSRC ICT Prioritisation Panel September 2021 Announced
Summary on Grant Application Form
Procedural geometry (PG) approaches use computer programs to create geometric models. For example, creating large scale 3D virtual cities instantly or searching aerospace procedural design spaces to optimise fuel consumption. Such powerful toolchains currently require users to first write complex computer programs, typically requiring expert knowledge of software engineering, linear algebra, and trigonometry. The lack of these obscure technical skills in the design and creative communities limits the number of people who can use PG to improve both their productivity, and the quantity, quality, and optimality of their designs.

PG has been well developed for niche domains, but its potential to improve productivity beyond expert users is neglected. As Esri Research notes "while CityEngine brings cutting edge procedural modelling to urban design professionals, the learning curve to create procedural rules means that we rely on large rule libraries to avoid the steep learning curve associated with writing procedural programs". This project will, for the first time, combine machine learning and high-context relational representations to make procedural geometry accessible to humans who do not want to code computer programs.

Procedural Geometry for Humans (PGH) will make PG accessible to a new audience of design and creative industries. It will develop novel deep learning approaches to geometric program generation in tandem with robust industry-facing prototype development; such prototypes will supply training data for machine learning (ML) whilst ensuring that our solutions are applicable to real-world problems.

Initially, we will design a representation for programmable geometry (WP1). Unlike existing approaches which store human-written code, the format will be optimized for automatic generation with ML from limited user given examples, rather than coding efficiency. From this platform PGH will engage with stakeholders to identify, define, and refine parametric applications of the representation; these will be leveraged to produce the first geometric procedural models dataset (WP2). The dataset and representation will be used to develop and train a novel deep neural network to generate PG by learning over the geometric contexts (WP3).

Finally, given this novel system we will demonstrate the utility and suitability of the PGH system for real world problems (WP4) by developing several PG-specific applications and demonstrators; these will apply the system to real-world problems which illustrate the power of PG (such as model fitting and the evaluation of massive-scale PGH models), and present the PGH system in a package suitable for application to industry and future research.

Industries which can benefit from PGH include building information modelling (estimated 2016 market size $3.16bn), video game development ($108bn in 2017), and movie post-production ($7.1bn in 2015).

The PI has extensive experience of procedural modelling in industry and research environments; this proposal will form the foundation of their future career. The postdoc hired for the project will have the opportunity to undertake world-class research, develop a publishing record, and benefit from the experience of the PI. This proposal combines the PI's deep knowledge of the procedural modelling industry with novel machine learning approaches to address the issues of generality and usability in PG. A publicly available implementation of PGH will provide a spring-board for industry to explore potential productivity gains while the creation of a PG dataset will provide the ML community to approach the issues that block industry from commercial opportunities in PG. Publications will explore ML for 3D geometry generation, the relationship between perceptual geometric relationships and the programs which generate them, procedural design in complex domains, and demonstrate the applicability of these new techniques to current industry problems.

Key Findings
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Potential use in non-academic contexts
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Impacts
Description This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Summary
Date Materialised
Sectors submitted by the Researcher
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Project URL:  
Further Information:  
Organisation Website: http://www.leeds.ac.uk