On Tuesday, January 22, the Metaverse Standards Forum (MSF) hosted a virtual town hall to explore a pressing question: Are Gaussian Splats (GS) ready for standardization? The event drew hundreds of participants and sparked a lively debate. The conversation has since continued at the “Gaussian Splats Town Hall Part 2” that was held on March 5, 2025, where Niantic, Cesium and Meta shared more details on their Gaussian Spat pipelines.
Gaussian Splats are an innovative method for representing 3D scenes using points (or “splats”) with attributes like color, opacity, and size. This technique enables highly realistic renderings with fewer data constraints compared to traditional mesh models.
The Forum’s Gaussian Splats Town Hall, co-hosted by MSF 3D Formats Interoperability Working Group co-chair Patrick Cozzi of Bentley Systems and Alexey Medvedev of Meta, who also serves as Khronos 3D Formats Working Group chair, was a dynamic, three-and-a-half-hour event that featured impressive demonstrations and persuasive arguments for GS’s future in 3D content creation and delivery. Notable contributors included Radiance Fields, The New York Times, Niantic, Meta, NVIDIA, PlayCanvas, Microsoft, Arrival.space, Waldek Technologies, Bentley Systems, Cesium, Hexagon, Esri, EARTHBRAIN, and JDULTRA.
The recording of Part 1 is now available, complete with bookmarks for easy navigation. As GS gains momentum, we also want to elevate the standardization discussion that emerged from the town hall and bring glTF stakeholders rapidly up to speed before the next event.What Does “Standardization Ready” mean?
Permanent link to this comic: https://xkcd.com/927/
Timing is everything in standardization. Premature efforts risk stifling innovation, while waiting too long can lead to fragmented proprietary technologies. The sweet spot is when a technology’s utility is proven, and the absence of standards starts to hinder growth.
This town hall focused on determining where Gaussian Splatting stands on this spectrum. Are we still in the early days of experimentation, or is the technology mature enough to benefit from standardization? And if it’s time, what should that process look like?
Growing Momentum for Gaussian Splatting
Presenters highlighted a rapid surge in GS adoption across industries, demonstrating its versatility and accelerating momentum. AJ Chavar from The New York Times illustrated how Gaussian Splats are enabling immersive, interactive storytelling with rich visual fidelity while maintaining minimal data overhead, crucial for fast-loading online content.
Nicholas Butko from Niantic showcased GS’s efficient rendering capabilities on mobile devices, emphasizing its ability to deliver smooth augmented reality experiences even on resource-constrained hardware. He also discussed Niantic’s Scaniverse app, a free and open-source tool that has significantly contributed to the growing adoption of GS. By enabling users to easily capture high-quality 3D scans using a mobile device and convert them into GS-based representations, Scaniverse has lowered the barrier to entry for creators and developers looking to experiment with and implement Gaussian Splats in their projects. Butko highlighted how Scaniverse’s streamlined workflow and intuitive interface have helped accelerate GS adoption across various industries, particularly in augmented reality applications.
Additionally, Michael Rubloff from Radiance Fields discussed how radiance field assets align with game engine physics, demonstrating integration with platforms like Unity and Unreal Engine. Nemanja Bartolovic and Ashish Singh from Meta (Cloud-AI, RL) highlighted their work optimizing GS for mobile GPU architectures, focusing on techniques to reduce memory bandwidth usage and enhance performance in resource-constrained environments like the Meta Quest. Nicolas Moenne-Loccoz from NVIDIA presented strategies for decoupling 3D Gaussians from specific rendering techniques, allowing for greater flexibility in how splats are visualized across different platforms and devices.These examples underscored GS’s ability to support high-quality visuals, efficient rendering across platforms, and adaptability for various industries, including journalism, augmented reality, and digital entertainment.
GS on the Web
The town hall also dove deep into GS web applications, highlighting its versatility for real-time use cases. Will Eastcott from PlayCanvas demonstrated how GS enhances real-time collaboration by enabling dynamic virtual spaces for education and remote work, where users can interact seamlessly within rich, interactive 3D environments. Cedric Guillemet from Microsoft showcased how GS improves the efficiency of rendering complex web-based scenes, making it possible to achieve high levels of detail without compromising performance. Thomas Richter-Trummer from Arrival.Space presented real-world applications where GS allowed for the scalable display of large 3D datasets directly in browsers, significantly reducing load times. Yoshiharu Sato from Waldek Technologies explored how GS supports immersive web applications, focusing on its ability to handle intricate data structures while maintaining smooth user experiences.
Despite these advancements, the presenters acknowledged challenges such as optimizing data compression techniques and ensuring consistent performance across different browsers. Ongoing collaboration within the community aims to address these hurdles and refine GS’s integration with web technologies.
Geospatial and Digital Twins
In geospatial and digital twin contexts, GS supports precise, high-resolution mapping, allowing for detailed representations of physical environments. Renaud Keriven from Bentley Systems demonstrated how GS could accurately capture and visualize complex infrastructure projects, including fine details like power lines, wires, or telecom towers, reducing the time needed for site assessments. The acquisition and processing of GS are more flexible than traditional photogrammetry. Data can be captured using anything from a simple smartphone video to georeferenced drone or aerial imagery. Processing ranges from a quick draft reconstruction to a more computationally expensive but high-quality, large-scale reconstruction.
Jason Sobotka from Cesium illustrated its capability in managing large-scale geospatial datasets, showcasing a real-time urban environment visualization that integrated dynamic traffic data seamlessly. This efficiency in data visualization makes it easier to interpret complex datasets, particularly in industries like urban planning and environmental monitoring.
Ladislav Horký from Hexagon discussed how meshes and point clouds could be used together, taking advantage of both the visual fidelity of a mesh and the fine structural details of a point cloud. His presentation demonstrated how point clouds can enable smoother performance and real-time data integration to enhance the responsiveness of digital twins.
Despite these benefits, challenges persist. Konrad Wenzel from Esri noted difficulties in standardizing coordinate systems across different platforms, while Jean-Philippe Pons pointed out the complexities in managing the vast amounts of data generated, both of which require further exploration and innovative solutions.
Are Gaussian Splats Standardization Ready?
The town hall revealed diverse perspectives on whether the industry is ready for GS standards. Pro-standardization voices advocated for leveraging existing frameworks like glTF to streamline GS adoption while maintaining flexibility. For instance, presenters from Cesium and NVIDIA highlighted how integrating GS with glTF could simplify workflows for digital twins and 3D mapping applications, enabling seamless data exchange across platforms.
Scott Simons from OGC offered a guiding principle during the global Q&A session: “The best way to standardize problems is really to find discrete pieces of those problems and solve them as those need to be solved.” In other words, stay niche. Simons suggested leveraging existing standards such as glTF that already address core aspects of GS use cases and building out extensions to address tailored user needs. This insight set the stage for a discussion on the right way to implement standards without overreaching.
Many participants suggested that glTF could play a pivotal role in future GS standardization efforts, potentially serving as a container or extension framework. Several speakers highlighted how 3D Tiles, a spatial data structure built on glTF, could enable streaming of massive GS-based scenes. Jason Sobotka from Cesium illustrated how 3D Tiles efficiently manages large-scale geospatial datasets, ensuring optimized delivery of high-resolution data in real-time applications. Konrad Wenzel from Esri reinforced this perspective, explaining how 3D Tiles’ hierarchical level-of-detail system allows GS to be integrated seamlessly into existing geospatial pipelines. Arrival.space also demonstrated how GS assets could be embedded within glTF files to enhance web-based 3D experiences without sacrificing performance or flexibility. These discussions further indicate that glTF, through its connection with 3D Tiles, may serve as a strong foundation for GS standardization, enabling interoperability and scalability across industries.
In addition, Niantic’s presentation introduced their open-source SPZ format for GS and proposed that SPZ could potentially become a glTF extension, following a path similar to Google’s Draco compression. They explained that SPZ is designed to efficiently store and transmit Gaussian Splats, making it a strong candidate for standardization. By leveraging glTF as an extension framework, Niantic suggested that SPZ could achieve rendering efficiencies, overcome challenges with interactivity, and build out metadata like view ranges to improve usability. Their proposal highlights how targeted extensions within established standards can provide practical solutions without overcomplicating the broader ecosystem.
Still, cautionary viewpoints emphasized the risk of stifling innovation if standards are imposed prematurely. Experts from Meta and PlayCanvas cautioned that GS is still evolving rapidly, with diverse experimental implementations that could be restricted by premature standardization.
Back on the pro-standardization camp, one should note that standards development is not a fast process. Getting started might not be such a bad idea and new technologies, which will inevitably arise, can always be incorporated.
Join the GS Conversation
The town hall demonstrated that the conversation around GS is far from over. Indeed, discussion continues on the event’s dedicated Discord, and the town hall will reconvene on March 5 at 9:30 AM PST.
If you’re interested in following the debate and weighing in, join the dedicated conversation on the Metaverse Standards Forum Discord. Members and non-members are welcome: by adding your voice, you can help ensure the industry hits the standardization sweet spot for this critical emerging technology.
Metaverse Standards Forum Gaussian Splats Town Hall Part 1 Speakers:
Hosts
Patrick Cozzi, Chief Platform Officer, Bentley SystemsAlexey Medvedev, AR Tech Lead at Meta and Chair of the Khronos 3D Formats Working Group
Presenters
Foundations
Michael Rubloff, Founder and Managing Editor, Radiance Fields
Aras Pranckevičius, Programmer
A.J. Chavar, Creative Strategist & Nick Bartzokas, Senior Software Engineer – R&D The New York Times
Momentum
Nicholas Butko, Senior Director, Engineering, Niantic, Inc.
Nemanja Bartolovic, Software Engineer & Ashish Singh, Product Management, Cloud-AI, RL-Meta
Nicolas Moenne-Loccoz, Principal Research Engineer, ADLR, NVIDIA
Web
Will Eastcott, CEO & Co-founder, PlayCanvas
Cedric Guillemet, Senior Software Development Engineer, Microsoft
Thomas Richter-Trummer, Co-Founder & CTO, Arrival.Space
Yoshiharu Sato, Founder & CTO, Waldek Technologies
Geospatial & Digital Twins
Renaud Keriven, Distinguished Engineer, Bentley Systems
Jason Sobotka, Software Engineer, Cesium
Ladislav Horký, CTO at Melown Technologies, Hexagon
Konrad Wenzel, Director, Stuttgart R&D Center and Jean-Philippe Pons, Sr. Principal Software Development Engineer, Esri
Takayuki Murayama, Senior Director & Talha Khalid, Global Development Manager, EARTHBRAIN
Emeric Beaufays, Geospatial and 3D Specialist, Founder & Software Engineer, JDULTRA
