Bring your implicit geometry to life: how to make currents, SDFs, and/or point clouds more surface-like ?

Details #

• Wednesday, november 19th 2025, 13:30
• Room : Metting room, Nautibus, 2nd floor

Bio #

Stephanie Wang is an unaffiliated researcher focusing on geometry, physics simulation, and computer graphics. She earned her Ph.D. in mathematics from UCLA under the supervision of Prof. Joseph Teran, and later held a postdoctoral position at UC San Diego with Prof. Albert Chern. Most recently, she worked as a Research Engineer at Adobe, where she contributed to meshing and design tools with a little touch of generative AI.

Abstract #

Implicit geometry representations such as currents, signed distance functions (SDFs), point clouds / Gaussian Spatial Representations (GSRs, or “gsplats”) are widely used to represent curved surfaces in computer graphics. While powerful and flexible, these representations often lack key properties of 2D manifolds, such as local parameterization and even integrability. This talk will present approaches for encouraging implicit representations to behave more surface-like, without converting them to meshes and therefore preserving their implicit strengths. We will discuss the geometric notion of what it means to be a 2D manifold, analyze how different implicit representations fall short of this ideal, and discuss optimization schemes that promote surface-like structure. Finally, we will consider the connection between these representations through the lens of geometric measure theory. This talk contains published and unpublished work.