Image Credit: 2025 Daiki Taniguchi
At SIGGRAPH 2024, Immersive Pavilion contributor Daiki Taniguchi offered visitors a new level of immersion in virtual environments with his project “GaussMR: Interactive Gaussian Splatting Sandbox With GPU Particles and Signed Distance Fields.” Capable of processing millions of Gaussians, it’s time to grab your notebooks and enjoy this deep dive into this work.
SIGGRAPH: Your project “GaussMR: Interactive Gaussian Splatting Sandbox With GPU Particles and Signed Distance Fields” looks to enhance XR applications. What inspired you to create this work? Was the main goal in mind to enhance XR applications?
Daiki Taniguchi (DT): While my expertise lies in GPU/GPGPU computing, I have consistently pursued research in achieving highly immersive XR experiences as an application domain for these technologies. Traditional triangle-based polygonal meshes have served us well, but their limitations have become particularly apparent in XR environments where visual fidelity is crucial. This observation drove me to investigate alternative 3D representations, including point clouds, voxels, and signed distance fields (SDFs).
The rapid advancement in Neural Radiance Fields (NeRF) and related technologies caught my attention as they aligned perfectly with my research interests. GaussMR was conceived not just as an enhancement to XR applications, but as a technical exploration combining my expertise in GPU computing with these emerging representations. The project implements numerous GPU-accelerated features, such as real-time SDF construction, collision detection simulation, and frustum culling for the Gaussian Splatting pipeline.
I find XR particularly fascinating as a platform where computer graphics and AI technologies converge. GPU computing serves as the foundational technology for both domains, and I saw an opportunity to showcase how these technologies can work together to create more compelling and interactive experiences.
SIGGRAPH: Whether you are working on mobile apps or working with Gaussian Splatting, there is a lot of trial and error that one may encounter. Can you discuss any roadblocks you faced while working on this project? How did you overcome them?
DT: While learning-based approaches are breaking new ground in computer graphics, they often face challenges in achieving real-time performance and interactivity. For instance, although there have been several papers attempting to add physical interactions to Gaussian Splatting, many of them are limited to offline simulations. Coming from a background in game development and real-time graphics, I believe that rather than trying to solve everything within the “AI” domain, it’s crucial to effectively integrate these technologies with existing computer graphics techniques. This philosophy forms the foundation of GaussMR’s key components.
One concrete example is how GaussMR reconstructs accurate SDFs from Gaussians in every frame. This not only required developing a highly parallelized, fast SDF construction algorithm but also presented another significant challenge. The trained Gaussian Splatting models often contain a considerable number of Gaussians that are either extremely small in scale or nearly invisible due to high transparency. Including all these Gaussians as SDF seeds would result in serious artifacts.
To address this issue, I developed a noise removal algorithm that examines the density of Gaussians in their local neighborhoods to determine whether a Gaussian might act as noise. Since I had already implemented spatial partitioning acceleration structures for fast collision detection simulation, the additional computational cost of this algorithm was negligible. This demonstrates how GaussMR effectively solves unique problems arising from learning-based methods using techniques from traditional computer graphics.
SIGGRAPH: Gaussian Splatting is a popular topic in computer graphics and interactive techniques. What inspired you to pursue this angle and submit this work to the SIGGRAPH 2024 Immersive Pavilion? What do you hope the SIGGRAPH 2024 audience took away, personally or professionally, from your Immersive Pavilion contribution?
DT: Actually, this marks my sixth acceptance to ACM SIGGRAPH’s Immersive Pavilion. I view the Immersive Pavilion as a unique fusion of academic conference and technology showcase. While technology is crucial (and, indeed, my greatest passion), it shouldn’t exist in isolation. Technology gains its true value when it leads to something novel, such as innovative experiences. I see the Immersive Pavilion as a bridge between these aspects, which is why I continue to submit my work. Interestingly, I’ve noticed a shift in the pavilion’s focus — around 2018, several submissions emphasized storytelling and artistic expression, but recent years have seen increased attention to underlying technical innovations. I believe this is a positive direction, as solid technical foundations are essential for advancing the exploration of innovative immersive experiences.
All my contributions have been guided by the principle that “innovative experiences are built on technological foundations.” While the XR industry has sometimes prioritized novelty and flashiness, my message to the SIGGRAPH community is clear: Let’s advance into new territories through the steady accumulation of robust technical knowledge.
SIGGRAPH: Technology in this field is rapidly evolving each day. How are you keeping up with the dynamic computer graphics industry? Are there any major plans to grow your work?
DT: I make it a priority to engage with primary sources — particularly reading research papers — and continuously learn from both emerging fields and established technologies with equal respect. I also find joy in witnessing the overall progress of our industry.
As for future plans, I’m not fixated on pursuing extensions of GaussMR. While Gaussian Splatting is an exciting field, we may see entirely new approaches emerge in the near future. However, just as GaussMR was built upon my various past research achievements, I’m certain that the technical insights gained from GaussMR, especially our approach to bridging AI and computer graphics through GPU computing, will shape my future research endeavors.
SIGGRAPH: Do you have any advice for someone who might be looking to submit their work to the Immersive Pavilion at SIGGRAPH 2025?
DT: The Immersive Pavilion is an exciting venue that combines the best aspects of academic conferences and tech showcases, offering a unique platform to transform your technical expertise into innovative experiences. I encourage you to submit your work with a pioneering spirit and your own distinct vision — this is your chance to shape the future with your own hands.
Feeling inspired? Use that inspiration to compose new realities at the SIGGRAPH 2025 Immersive Pavilion. Not sure if the Immersive Pavilion is the correct program for your work? Not a problem! Explore all of the programs that are accepting work for SIGGRAPH 2025.
Daiki Taniguchi is a research engineer specializing in GPU/GPGPU computing, with extensive experience in real-time graphics, interactive systems, and machine learning. His work has been accepted six times to the ACM SIGGRAPH Immersive Pavilion, demonstrating the convergence of high-performance computing and immersive technologies. View Daiki’s X and LinkedIn accounts.
