Silvia Sellán
@silviasellan.bsky.social
Assistant Professor of Computer Graphics and Geometry Processing at Columbia University https://www.cs.columbia.edu/~silviasellan/
I know language evolves and we shouldn't be prescriptive and all that but.... please please please what is with people saying "else" instead of "otherwise"? We talk a lot about coding using natural language but this reverse effect of changing language to imitate code is horrifying.
The talk I presented at #SIGGRAPH2026 is now available on YouTube, for anyone who couldn’t attend or would like to revisit it :) youtu.be/OFJYBUArVPc?...
Dual Contouring of Signed Distance Data - SIGGRAPH 2026 [Technical paper presentation]
YouTube video by Xiana Carrera
youtu.be
Xiana Carrera@xianacarrera.bsky.social · 2mo ago1/6 Excited to share our #SIGGRAPH2026 paper, Dual Contouring of Signed Distance Data! We propose a new method for reconstructing sharp meshes from discrete samples of signed distance functions (SDFs) that does not require gradients. Project page: gatc.cs.columbia.edu/projects/dua...
Some initial thoughts, and a complicated mix of feelings. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not.
I was asked about a small breakdown for the 3D anime ramen and realized my breakdown thread was on the cursed site from which I deleted my account 😅 So I'm gonna post a small 🧵 here
Ah, this is great. I love it. I love the Tolstoy quote too. I like every part of this
Claas Voelcker@cvoelcker.bsky.social · 6d agoThis is a robot failing to grasp a ball. Almost every robot lab produces clips like this daily… and almost all of them get thrown away. This is the most abundant but underused resource in robot learning. We’re collecting all of it now as “OopsieData”, please join us at oopsie-data.com! (1/13)
This is a robot failing to grasp a ball. Almost every robot lab produces clips like this daily… and almost all of them get thrown away. This is the most abundant but underused resource in robot learning. We’re collecting all of it now as “OopsieData”, please join us at oopsie-data.com! (1/13)
One of my favorite papers this year, I am super impressed by Nicole's work!
Nicole Feng@nicole--f.bsky.social · last wk.At #SIGGRAPH2026, I presented "points as tori" (PAT) for fast pointwise evaluation of signed distance for point clouds. Some code (Python w/ C++ extensions) and demo up: github.com/nzfeng/point... Contributions from people/LLMs welcome. Explainer thread below for the project: 🧵
People not in tech have been interested in what happened with the Open AI agent/Hugging Face hack. So, I "Explain it Like I'm 5"-ed it and made a cartoon in 8 panels. Apologies for some anthro-ing. 🫤 m-mitchell.com/HF-hack-cart... 1/8
#SIGGRAPH2026 attendees: I made a better webpage for the Technical Papers schedule. You can search, filter on keywords, and scroll through linearly (rather than opening a zillion folds) + add to calendar: keenancrane.github.io/siggraph-pap... Can also download a flat JSON & CSV of talks & times.
Next week at #SIGGRAPH 2026, we'll present "Locality-Aware Automatic Differentiation on the GPU for Mesh-Based Computations," with Rahul Goel, Jonathan Ragan-Kelley, and @justinmsolomon.bsky.social . Talk: Mon., July 20, 2:50 PM, Room 403-B Paper: dl.acm.org/doi/10.1145/... More below (1/7):
Excited to share our latest work on discrete SDF reconstruction with no gradients required! Please come to Xiana's talk in LA :)
Xiana Carrera@xianacarrera.bsky.social · 2mo ago1/6 Excited to share our #SIGGRAPH2026 paper, Dual Contouring of Signed Distance Data! We propose a new method for reconstructing sharp meshes from discrete samples of signed distance functions (SDFs) that does not require gradients. Project page: gatc.cs.columbia.edu/projects/dua...
Ever since I have worked in mesh reconstruction from SDFs, the general consensus was "to get sharp features, you need access to the SDF gradient (e.g., with dual contouring)". Amazingly, Xiana shows that this is not true, and one can reconstruct sharp features from SDF data alone. Crazy!
Xiana Carrera@xianacarrera.bsky.social · 2mo ago1/6 Excited to share our #SIGGRAPH2026 paper, Dual Contouring of Signed Distance Data! We propose a new method for reconstructing sharp meshes from discrete samples of signed distance functions (SDFs) that does not require gradients. Project page: gatc.cs.columbia.edu/projects/dua...
6/6 Huge thanks to my wonderful coauthors @ninnnnnglala.bsky.social, @topher-batty.mastodon.acm.org.ap.brid.gy , @odedstein.bsky.social, and my amazing PhD advisor @silviasellan.bsky.social. Check our paper here: gatc.cs.columbia.edu/assets/pdf/d... (Code release coming soon!)
5/6 The results are sharper and more accurate reconstructions from sampled SDF data, especially at medium-to-high grid resolutions, with applications to narrow-band meshing and swept-volume reconstruction.
4/6 We then incorporate these constraints into local, parallelizable quadratic optimization problems to refine mesh vertex positions, and iteratively update the intersections between edges and the isosurface, as well as their corresponding gradients, to recover sharper geometry.
3/6 Our approach revisits the classic Dual Contouring, but removes the need for exact gradients as input. To do so, we leverage the fact that each SDF sample gives distance information in the form of a spherical constraint that the unknown surface should be tangent to.
2/6 SDFs are widely used in modeling, simulation, manufacturing, and graphics, but many downstream tasks still need explicit surface meshes. From grid samples alone, sharp reconstruction is hard, and existing methods often smooth features or require extra information.
1/6 Excited to share our #SIGGRAPH2026 paper, Dual Contouring of Signed Distance Data! We propose a new method for reconstructing sharp meshes from discrete samples of signed distance functions (SDFs) that does not require gradients. Project page: gatc.cs.columbia.edu/projects/dua...
kamathematics.wordpress.com/2026/05/27/m...
Making a talk, without and with AI
Some of the discussion online has been about how not to use AI in making academic talks (see, e.g., this post by Jessica Hullman). A junior researcher asked my opinion on using AI to help make slid…
kamathematics.wordpress.com
We are excited to announce the 2026 cohort of Rising Stars in Computer Graphics 🌟 These 9 young researchers will participate in a workshop at Siggraph 2026, about the wide range of career trajectories in graphics. Thank you to all applicants and to our selection committee for all the hard work.
Proud of my students for their research, which for the first time located the descendants of those enslaved by Columbia University's founders. www.columbiaspectator.com/news/2026/05...
Student researchers identify living descendant of person enslaved by Columbia founder
For the first time since the Columbia University and Slavery seminar began in 2014, students have identified a living descendant of someone enslaved by a Columbia affiliate.
columbiaspectator.com
May 3, 1968 the #GratefulDead got smuggled past student barricades into the quad at Columbia University in NYC to play a stealth concert for the students who were protesting youtu.be/WFtQr_W3opM?... 📷 Rosie McGee
I am continually asked if VFMs "understand" classical mechanics and my benchmark for this is the venerable newton's cradle. Let's take a look at progress over the last year using Google's Veo models -- a 🧵
At #CHI2026 we are also presenting "StepDance: A Toolkit for Redesigning CNC Machines Using Physical Metaphors" Curious how a machine-building novice like me can make a sketch-to-print machine? checkout the talk on Friday >> programs.sigchi.org/chi/2026/pro... or see thread...
