Silvia Sellan

Research Abstract:

I use geometric and statistical tools to study the uncertainty in Computer Graphics tasks. My research lays the mathematical foundation for my vision of a world in which 3D graphics are propelled beyond their traditional entertainment applications and into decision-making in critical industries like medicine and engineering. Despite its origins as a field in industrial design and medicine, the economic incentives of the past decades have made it such that the vast majority of Computer Graphics research has focused on the needs of film, video game and social media companies. This influence has affected which problems are studied and which algorithmic qualities are valued: we have enabled artists to freely create high-quality 3D geometry to be shown to audiences in theatre screens, produced interactive realistic-looking simulations in less than the time it takes a gaming console to render a single frame, and captured the geometry of real-world 3D products for use in advertising. These techniques have become so well established that they are routinely used in fields far beyond our own like medicine, security and engineering. However, this risks turning Computer Graphics research into a victim of its own success, as different application realms introduce a novel set of geometric inputs and algorithmic priorities. At best, relying on the same skewed research incentives of the past decades limits the application realm of Computer Graphics; at worst, it sets the stage for catastrophic consequences as colleagues in other fields increasingly use our algorithms to train their models for more sensitive applications. As academics, we are in a privileged position to resist the impulses of industry and steer research in the service of society. In Graphics, this means designing robust algorithms that accept a diverse set of inputs beyond high quality artist-designed triangle meshes, and quantifying the uncertainty of our predictions in a way that can be communicated to (human or artificial) decision-makers. To solve these challenges, I propose understanding Computer Graphics tasks through a statistical lens: interpreting our algorithmic decisions as statistical priors, inputs as observations and outputs as inferences drawn from posterior distributions. My previous and ongoing work apply this perspective to different geometric capture and modeling tasks and set the stage for future work that carries it further down the Graphics pipeline and into specific applications.

Bio:

Silvia is a fifth year PhD student at the University of Toronto, advised by Alec Jacobson. In the course of her PhD, she has led ten first-author publications at top-tier Computer Graphics and Vision venues and participated in several others. She is a Canada Vanier Doctoral Scholar, an Adobe Research Fellow and the winner of the 2021 University of Toronto Arts & Science Dean’s Doctoral Excellence Scholarship. She has interned twice at Adobe Research and twice at the Fields Institute of Mathematics. She is also a founder and organizer of the Toronto Geometry Colloquium and a member of WiGRAPH. She is currently looking to survey faculty positions to start in Fall 2024.