It is quite difficult to write code that extracts every ounce of performance out of a modern GPU. Doing so requires a deep understanding of the chip architecture and often lots of careful benchmarking and experimentation to see how best to map your workload onto the hardware. LLMs are exceptionally good at this type of software task, though, and these 3DGS speed ups show how much performance is sometimes sitting on the table!
I love too how the knowledge here is being shared openly and in a way will lead to a lot of rapid adoption (and widespread benefit for those who use this rendering stack). Thanks @gkopanas !
Is it possible to make 3D Gaussian Splatting even faster?
Short answer YES!
Enjoy a 10.5x optimization on forward/backward cuda kernels over the published 3DGS ones!
grgkopanas.github.io/3dgs_cu…
5
120
Michael Broxton retweeted
Amazing to see deployed soldiers share a moment with their kids through Beam. Huge thanks to @USArmy @The_USO for helping to make this possible.
🇺🇸 For a deployed service member, missing family moments adds up. Baby’s first step. The first performance. Sharing a laugh.
In honor of National Day of the Deployed, we’re excited to partner with @USArmy and @The_USO to bring Google Beam to 20 locations so deployed service members can experience lifelike, 3D video connections with their loved ones.
🎥 Watch the docuseries below
Michael Broxton retweeted
Picture stepping into a live 3D broadcast or 3D video call and looking around freely. That is the promise of dynamic NVS.
But multi-view camera setups are fixed so parts stay hidden. Looking anywhere means recovering them from memory, which current stateful and stateless models cannot do in real-time over long-horizons.
Until now. Introducing NSTM.
⚡️Real-time inference + 🧠minute-long memory.
Michael Broxton retweeted
📢 Google's Project Beam (formerly Project Starline) is looking for Student Researchers!
We seek students interested in real-time high-realism 3D reconstruction, dynamic 3D, relighting, materials, novel view synthesis, diffusion, video models, transformers, and more.
🧵1/7
Michael Broxton retweeted
Excited to present "LVT: Large-Scale Scene Reconstruction via Local View Transformers" at @SIGGRAPHAsia 2025, Hong Kong 🏙️
💡Linear-complexity, large-scale scene reconstruction in a single inference pass!
⚡Fast-forward: Mon 9am HKT (Hall 3F)
🗣️ Technical Presentation: Thu 1:10pm HKT (S426)
🔗 Project Page: toobaimt.github.io/lvt/
📄 Paper: dl.acm.org/doi/10.1145/37573…
Work done during my internship at @Google Beam, with my amazing collaborators John Flynn, @lucyrchai, Kathryn Heal, @XuanLuo14, and Jungyeon Park ✨
#ACMSIGGRAPH #SIGGRAPHAsia2025 #3DGS
Michael Broxton retweeted
Project Starline is going from research to reality. It will now be called Google Beam! Our advanced AI doesn't just connect you—it feels like it teleports you in stunning, lifelike 3D.
The first devices from @HP launch later this year for early customers!
📢 Google's Project Starline is looking for Student Researchers!
Our team seeks students interested in real-time, high-realism human 3D modeling, relighting, materials reconstruction, novel view synthesis, diffusion models, transformers, and more.
🧵Look below for details!
Previous student researchers have published work they started with our team at CVPR, SIGGRAPH, ECCV/ICCV, and contributed to innovative products including Project Starline, Google Jump VR Camera, Lens Blur, Cardboard Camera, Google Clips, Photo Tours and the Ceres Solver. Our latest papers include Quark, Inpaint3D, Nerfies and HyperNeRF, DeepView, Light Field Video, and A Dark Flash Normal Camera.
1
3
1,136
Michael Broxton retweeted
Thrilled to announce our paper, Quark: Real-time, High-Resolution, and Generalized Neural View Synthesis, has won Best Paper at SIGGRAPH Asia 2024! 🎉
Grateful for our team’s exceptional collaboration. 💪
Explore our work: quark-3d.github.io
#SIGGRAPHAsia2024
Today we are pleased to present Quark, our new method for real-time, high-resolution, and generalized neural view synthesis! We will be giving our talk at #SIGGRAPHAsia2024 in Hall B5 (2) at 10:45am.
🚀 Quark renders 1080p @ 30fps on a single A100 GPU
🌎 Generalizes across diverse scenes & datasets
🏆 Achieves quality rivaling even offline methods
🔗 Check it out! quark-3d.github.io
Michael Broxton retweeted
Excited to share our latest paper: Quark – Real-time, high-resolution, and generalized neural view synthesis! 🚀
🎥 1080p @ 30fps
🌍 Generalizes across diverse scenes & datasets
🏆 Achieves real-time quality rivaling even offline methods
🔗 Project page: quark-3d.github.io/
Excited about the release of our method that performs 3D inpainting guided by a generative prior! Congratulations @jane_h_wu, Kira, Lynn, @poolio, @PeterHedman3, and @danbgoldman!!
This Post is from an account that no longer exists. Learn more
2
384
Interested in Project Starline (blog.google/technology/resea…) and/or generative imagery and media? My team at Google is looking for interns for summer 2023!
Application link: bit.ly/3XY4Yim
Email project-starline-intern-summer-2023@google.com to tell us that you applied!
Michael Broxton retweeted
Há um grupo de investidores privados que quer construir um novo aeroporto, sem investimento público e de forma faseada na região de Santarém.
Leia mais, em exclusivo no Expresso: swki.me/wvVY8WMV
Michael Broxton retweeted
Project Starline is publishing a technical paper about the system at Siggraph Asia 2021 in December. @jlawrence176 and @danbgoldman also made a terrific presentation video, check it out!
paper: research.google/pubs/pub5090…
presentation video: youtube.com/watch?v=kDgToq5a…
If you are a computer vision or computer graphics PhD student and interested in a summer internship, please consider applying to my group at Google – project Starline!
My group at Google, Project Starline, is looking for interns for summer 2022! Please consider applying!
Application link: bit.ly/3HG60bc
Project Starline announcement: blog.google/technology/resea…
