About me

Hi! I am a second-year Computer Science PhD student at the University of Toronto, advised by Alec Jacobson and working mainly in Computer Graphics and Geometry Processing. I completed a B.Sc. in Mathematics and a B.Sc. in Physics at the University of Oviedo. I have interned twice at Adobe Research under the mentorship of Noam Aigerman and twice at the Fields Institute of Mathematics. You can click here to download my full resume. The best way to contact me is to email me (preferrably in English or Spanish) at sgsellan (at) cs.toronto.edu, but you can also message me on twitter or follow me to stay up to date on my research (combined with some random thoughts every now and then). I encourage you to look at the bottom of this website for some of the projects I am currently working on or interested in chatting about!

Publications

	Opening and Closing Surfaces Silvia Sellán, Jacob Kesten, Ang Yan Sheng, Alec Jacobson ACM Transactions on Graphics (SIGGRAPH Asia), 2020 [Paper] [Paper (low res)] [Video] [Project Page] [Code (coming soon)]
	Developability of Heightfields via Rank Minimization Silvia Sellán, Noam Aigerman, Alec Jacobson ACM Transactions on Graphics (SIGGRAPH), 2020 [Paper] [Paper (low res)] [Code] [Project Page] [Talk (English & Spanish CC)] U.S. Patent pending
	Solid Geometry Processing on Deconstructed Domains Silvia Sellán, Herng Yi Cheng, Yuming Ma, Mitchell Dembowski, Alec Jacobson Computer Graphics Forum (Eurographics), 2019 [Paper] [Paper (low res)] [Arxiv] [Code] [Project Page] [Talk]

Projects and Research Interests

Here are some of the things that are on my mind or that I've been working on lately. However, this is by no means an exhaustive list of my interests and I'll be more than happy to accept new ideas/projects.

Discrete developable surfaces

A developable surface is a surface that is locally isometric to the plane. In practice, what this means is that one would be able to construct this surface using bent (but not stretched) sheets of materials like paper or metal. These surfaces are omnipresent in our world: the pages of a book, origami structures, the hulls of ships and buildings like the Los Angeles Walt Disney Concert Hall or the Bilbao Guggenheim Museum.

In the discrete world, even defining developability is tricky. Triangles are trivially isometric to the plane, so any triangular mesh is strictly speaking piecewise developable. Many fascinating works from the past years consider novel definitions of developability. While groundbreaking, some are too dependent on the surface's discretization, or can only consider a limited and/or prescribed number of developable patches. Our goal is to find a definition of (piecewise) developability that is as discretization-agnostic as possible and imposes no restrictions on the location or number of developable patches. This notion would need to be suitable to be optimized for in a mathematically tractable way (ideally, as a convex energy).

Our latest paper, presented at SIGGRAPH 2020, comes very close to doing just that, but with one important caveat: it is restricted to work only on heightfields. The work in this paper was conducted during my 2019 internship at Adobe Research in San Francisco, mentored by Adobe Research Scientist Noam Aigerman and my PhD advisor Alec Jacobson. Currently, I am interested in ways in which we can extend our work to general surfaces in 3D space. I am always happy and ready to chat and share ideas that go in this direction.

Discrete differential geometry on CSG-constructed shapes

Constructive Solid Geometry is a way of building shapes as basic set operations (unions, intersections and substractions) of very simple ones called primitives. While this is a very useful tool for constructing very complex final 3D objects, one runs into trouble when attempting to define differential operations on the final solid shape. Usually, this means having to obtain a surface mesh of the final shape by running boolean operations on the primitives, and then tetrahedralizing this final complex domain with all the possible issues which that entails.

We are looking for ways to define these operators by tetrahedralizing the primitive shapes, then using only the information provided by these triangulations, thus removing the need for complex remeshing of the final shape. In our most recent paper, published in Computer Graphics Forum and presented at SGP 2019, we solve this problem for some equations and for the specific case of mesh unions. The collaboration that led to this paper was a joint work supervised by University of Toronto professor Alec Jacobson, and supported by the Fields Institute for Research in Mathematical Sciences. I am excited to extend our work to more general set operations, like intersections and substractions, and would love to chat and share ideas in this regard (as it turns out, it isn't as trivial as it sounds!)

Smoothing geometric flows inspired by morphological operations

Morphological operations are a well-studied area of mathematics which has lately found interesting uses in the realm of Computer Graphics, from pre-processing shapes for 3D printing and CNC milling to overall use as a smoothing tool. The traditional pipeline for computing these operations on a three-dimensional surface implies transforming it into a very fine 3D binary image, to then apply the standard definition of the operation as a binary, and to finally re-transform the resulting binary into a 3D surface. This procedure is lossy in information and has intrinsical issues when dealing with other data defined on the surface. We are looking for ways to define these operations as geometric flows directly on the surface so as to avoid the aforementioned problems.

In our SIGGRAPH Asia paper, we do precisely this but for the specific case of the opening and closing operations and spherical structuring elements. Ideally, we'd like to define any morphological operation this way regardless of the chosen structuring element, and do it in a way that handles holes and self-intersections in the surface. Do feel free to contact me (see the top of this website) to learn more about our progress or to submit ideas or suggest collaborations.

Questions to ask as a prospective graduate student

In the spring of 2019 I faced a hard decision on which lab to join for my PhD. Something that really helped with making a choice was asking a consistent set of questions to every prospective advisor. I feel like this may be useful to students in similar situations, so I've decided to share this set of questions with the world here.

EDIT: My labmate Towaki Takikawa wrote this amazing overview of Graduate School in Canada, which I highly recommend if you're considering different schools to apply to!

A SIGGRAPH first-timer's log

In the summer of 2018, while I was just an undergraduate student, I attended SIGGRAPH for the first time. This was my first time ever going to such a large-scale conference, and it impressed me so much I decided to write a log with my daily thoughts. It ended up being the experience that convinced me into studying Computer Graphics and pursuing my current PhD, and I still think of it as a motivation every now and then. Anyway, I've decided to archive this log here.