Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Launch Supplement

Université de Montréal

$12,500.00

Practical Neural Geometry Processing

FY 2024-25

Summary
Summary The world surrounding us is three-dimensional, and geometric problems related to it pervade all areas of science. Tackling these problems computationally, using artificial intelligence (AI) requires successful application of the current trends in machine learning (namely, deep learning) to 3D geometry processing. Currently, application of AI to such tasks usually resorts to techniques from 2D computer vision that operate on pixel-like grids, instead of actual 3D shapes and models. This is due to specific computational difficulties 3D geometry poses (more complex representations; mathematically-difficult tasks and objectives). Without directly operating over 3D objects, we cannot expect to conquer tasks that require real 3D understanding and interaction, without relying on visual perception and pixel-like grids. For instance, analyzing and constructing operating motorized structures that have many occluded, interlocking parts at different scale lengths, cannot be achieved through vision. This proposal aims to provide core techniques that can enable the application of deep learning to real-world 3D geometry processing. These solutions can be thought of as a "swiss army knife", comprising several tools that can be successfully employed in most practical scenarios. Namely, in the next 5 years I will focus on 3 core computational tasks: 1) Modifying and manipulating 3D shapes through deep learning. 2) Learning over 3D shapes. 3) Matching and mapping between 3D shapes. These core methods will be devised with focus on broad real-world applicability, mandating several key properties: generality, ensuring they do not take any task-specific design choices, and can be applied to arbitrary tasks; robustness to noisy and partial data; efficiency in performance and memory footprint; and high accuracy of the output of the method, a critical necessity for many 3D geometry applications. In order to achieve these ambitious goals, I will use my expertise in geometry processing and its intersection with deep learning, to devise theoretically-grounded new representations and approaches that tightly-integrate 3D geometry and machine learning. On its own, each of these three core components will significantly boost AI capabilities for 3D tasks, while successfully tackling all three will synergistically provide far-reaching complete solutions to critical problems. As a blue-sky example, these tools could potentially enable designing a system for predicting healthy organ growth: a system could receive a 3D scan of a newly-born infant's heart; it could then analyze it at extremely-high resolution using the technology provided by track 2, compare and match it to scans of other specimen from a database (using track 3), and then use learned priors to simulate its growth and development, "morphing" it to its final adult state, using track 1.
Program
Discovery Launch Supplement
Location
Quebec, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
DGECR-2024-00439
Name nom
Aigerman, Noam
Department département
Department of Computer Science and Operations Research
Organization id
63
Institution établissement
Université de Montréal
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
12500
Program id
DGECR
Program name en
Discovery Launch Supplement
Program name fr
Supplément Tremplin vers la découverte
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1207
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Information, computer and communication technologies
Area of application fr
Technologie de l'information, des ordinateurs et des réseaux de communications
Research subject code
2707
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Graphics
Research subject fr
Infographie
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Practical Neural Geometry Processing
Keywords
geometry processing; deep learning; 3d geometry; deformation; mapping; shape analysis
Application summary
Summary The world surrounding us is three-dimensional, and geometric problems related to it pervade all areas of science. Tackling these problems computationally, using artificial intelligence (AI) requires successful application of the current trends in machine learning (namely, deep learning) to 3D geometry processing. Currently, application of AI to such tasks usually resorts to techniques from 2D computer vision that operate on pixel-like grids, instead of actual 3D shapes and models. This is due to specific computational difficulties 3D geometry poses (more complex representations; mathematically-difficult tasks and objectives). Without directly operating over 3D objects, we cannot expect to conquer tasks that require real 3D understanding and interaction, without relying on visual perception and pixel-like grids. For instance, analyzing and constructing operating motorized structures that have many occluded, interlocking parts at different scale lengths, cannot be achieved through vision. This proposal aims to provide core techniques that can enable the application of deep learning to real-world 3D geometry processing. These solutions can be thought of as a "swiss army knife", comprising several tools that can be successfully employed in most practical scenarios. Namely, in the next 5 years I will focus on 3 core computational tasks: 1) Modifying and manipulating 3D shapes through deep learning. 2) Learning over 3D shapes. 3) Matching and mapping between 3D shapes. These core methods will be devised with focus on broad real-world applicability, mandating several key properties: generality, ensuring they do not take any task-specific design choices, and can be applied to arbitrary tasks; robustness to noisy and partial data; efficiency in performance and memory footprint; and high accuracy of the output of the method, a critical necessity for many 3D geometry applications. In order to achieve these ambitious goals, I will use my expertise in geometry processing and its intersection with deep learning, to devise theoretically-grounded new representations and approaches that tightly-integrate 3D geometry and machine learning. On its own, each of these three core components will significantly boost AI capabilities for 3D tasks, while successfully tackling all three will synergistically provide far-reaching complete solutions to critical problems. As a blue-sky example, these tools could potentially enable designing a system for predicting healthy organ growth: a system could receive a 3D scan of a newly-born infant's heart; it could then analyze it at extremely-high resolution using the technology provided by track 2, compare and match it to scans of other specimen from a database (using track 3), and then use learned priors to simulate its growth and development, "morphing" it to its final adult state, using track 1.