SIGGRAPH 2024
This year I had the fortunate opportunity to attend SIGGRAPH 2024 with Samsung sponsoring my travel. It was my first time attending, and my second convention (GDC in March 2024 being the first) as a professional.
Back in high school and middle school I had always known about SIGGRAPH, since I had a big interest in computer graphics. I was always learning how to use modeling tools like Maya and Blender, and was even subscribed to 3D World for a period of time.
Now my interests mostly lie in the technical domain, so that was my main focus this year. Just wanted to write up a summary of things that I learned and experienced for future me to look back on.
Technical Papers and Talks
WARP: Differentiable Spatial Computing for Python
This talk was more for the simulation and scientific industry, but was cool nonetheless. WARP is a Python framework for high performance GPU simulation and visualization. Essentially you write both your driving code and kernel code in Python (using special function decorators) and the WARP framework will generate and compile C++ and CUDA to then execute your program. I don't work in this field but it was pretty neat to see how easily you could get complex simulations and a fancy visualization put together.
CG in Japan
This session was a collection of talks from various Scientists in Japan with not really a common theme tying them together. Yuta Noma (PhD student) talked about some of his works regarding meshes and surface-filling curves. Professor Yoshinori Dobashi presented a project on Efficient Visualization of Light Pollution for the Night Sky which I thought was pretty neat.
HyperPose
Alex Bereznyak of Activision presented HyperPose, which is a brand new animation system that attempts to address the various problems that plague current modern approaches to realistic animation. Even though I had 0 knowledge in this field besides basic things like blend shapes and keyframes, this was definitely a fantastic talk and Alex was an amazing presented. You could tell he was really passionate about this subject.
Opening Keynote with Mark Sagar: Beyond the Illusion of Life
This talk was weird, but in a really interesting and cool way. Dr. Sagar started off with the premise that realistic animation is what helps make something feel "real" (or human? don't remember exactly). He went on to talk about some of his research and past accomplishments like the impressive facial animation work on movies like Avatar. Everything from innovations in motion capture to fully simulating muscles and tissues of the face. All of this culminated in their demo called BabyX: Which is essentially a fully simulated (physical and neurological) child. I was honestly pretty impressed (and a bit creeped out) by the work. Very few technical details mentioned other than they built simulation mechanisms for every aspect of a human.
Differentiable Rendering
There was a series of talks about this relatively new topic in (mostly) offline rendering. Differentiable rendering is writing your renderer s.t. you can take derivatives of your rendering functions with respect to parameters like models, camera view, light positions, etc. With derivatives you can now formulate various optimization problems around your renderer. The most interesting one I saw was a paper on optimizing light arrangements in a 3D scene via "differentiable light tracing". One of the applications was taking a fully baked scene from Quake III arena, and reconstructing the original light positions (based on the baked light map) into something that could be used for modern real time lighting. Something like RTX Remix could really use this technique, since often times we have no idea where the lights were placed after the fact.
Advanced in Real-Time Rendering in Games
This in some ways was almost the main event for me. Highly recommended by co-workers, it was a series of back-to-back presentations of production-focused game technologies. Some of the topics went a bit over my head since I don't work in those areas (yet), but I found a few of them very useful and interesting:
Neural Light Grid: Modernizing Irradiance Volumes with Machine Learning by Michał Iwanicki (Activision)
- This as a whole was a refreshing talk about a practical approach to using ML in games. A lot of the recent hype around ML in games is just that -- hype. Michał's talk other than the technical stuff was pretty much a journal on what experimenting (and sometimes failing) with ML research in games is like, and where it makes sense to apply it.
Variable Rate Shading with Visibility Buffer Rendering by John Hable (Visible Threshold)
- I felt like this was more of a talk on promoting applications of a Visibility Buffer, than VRS. Either way, it had a really great explanation and visualization of quad overdraw in the rasterizer than is scarcely found on the internet. Definitely will be research Visibility Buffers after this one.
Shipping Dynamic Global Illumination in Frostbite by Diede Apers (EA | Frostbite)
- Cool talk on using surfels to help with the GI problem in production. I don't know a ton about GI at the moment but the performance results seemed quite impressive.
Hair
Real-time Physically Guided Hair Interpolation by Jerry Hsu and Real-time Hair Rendering With Hair Meshes by Gaurav Bhokare
- Grouped together because they are under the same research group, but some really great easy to follow talks on real-time hair rendering that goes beyond just hair cards. The WebGL demo with hair sliders was really a great touch.
Stylized Production
Real-time Refraction Shader for Animation by Antoine Domon
- Short but sweet talk from Animal Logic talking about how they implement a refraction effect specifically of the eyes in the viewport, allowing artists to have better foresight into the final render. Basic idea is raytracing the eye with a (straight on?) camera view using NVIDIA OptiX, and doing some performant transfers over to OpenGL where they project back into the viewport view in a fragment shader.
Dynamic Screen Space Textures for Coherent Stylization by Brent Burley
- Talk about how they achieved a watercolor effect in Disney's Wish. Much more involved in getting a pleasing result than you might think. Honestly I found the end result to be quite subtle for how much work was put into it (maybe it was the compression) but the technical details were great nonetheless.
Character Animation: 2D, 3D, Robots
Text-guided Synthesis of Crowd Animation by Xuebo Ji
- Most of the time crowds rely on artists to finely tune their behavior. The insight this paper presents is that environments usually contain rich priors like crosswalks and other landmark features. The authors use a labeled environment map along with an LLM processed description of the crowd groups to generate maps (start/goal "agent distribution" and a velocity field) for guiding the simulator.
Interactive Design of Stylized Walking Gaits for Robotic Characters by Michael A. Hopkins
- Gait editor for robotics that works in real-time with physical robots (and simulated), very cool! Specific emphasis on design features (artist-oriented) being married to realistic physical motion planning (engineering-oriented). Robust to mild disturbances like pushes and supports asymmetric gaits (limping for example).
Mobile Graphics
Moving HypeHype towards Physically Based Rendering by Sebastian Aaltonen
- Stuff from Sebastian is always a joy to learn from. His Twitter is basically a goldmine of info, especially for mobile rendering. Given my current job in mobile graphics, it was a very useful talk! Mostly focused on low-end Android devices, which we often forgot is a big share (if not the majority) of devices out there.
Electronic Theater
About 2 hours of short animated films. Some really inspiring stuff, and all I know is that the French are onto something.
Denver Downtown
I definitely did not expect it to be this hot in Denver in July. It was even worse than what we were getting in the Bay this year. Food in downtown Denver ranges from alright to honestly pretty bad. But there is a Giordanos which I was pleasantly surprised by.
Miscellaneous Thoughts about the conference
I had always gotten the impression that SIGGRAPH was sort of a balance between less-technical and technical aspects of graphics, but it definitely felt like it was heavy on the technical side. According to a co-worker the amount of art related things has decreased in the many years he had attended.
The exhibition area was small, and a lot more company/product oriented than GDC, which seemed to have a lot of booths who were just there to chill and talk.
Overall the conference was smaller (or at least felt smaller) than something like GDC. Not necessarily a downside.