3D Gaussian Splatting

Description

3D Gaussian Splatting (3DGS) is a novel scene representation and rendering technique that models a three-dimensional scene as a collection of 3D Gaussian primitives, each with a position, covariance, opacity, and colour (often represented via spherical harmonics). Unlike neural radiance fields (NeRF), which use a neural network to represent the scene, 3DGS uses explicit, differentiable Gaussian primitives that can be rasterized efficiently, enabling real-time rendering on commodity GPUs. In 3GPP Release 19, 3DGS is studied under TS 26.956, which addresses media handling and rendering for immersive services, including extended reality (XR) and volumetric video.

Architecturally, 3DGS fits into the media pipeline as a representation format for volumetric content. The scene is captured from multiple viewpoints, and a training process optimizes the Gaussian parameters to match the input images. The resulting set of Gaussians is then stored or streamed, and a renderer projects them onto the screen using a splatting rasterizer. In the 3GPP context, this involves defining how 3DGS content is packaged, transported, and decoded within the media framework. TS 26.956 specifies the media handling aspects, including support for dynamic scenes, view-dependent effects, and integration with existing 3GPP media services such as 5G Media Streaming and MMS.

Key components of 3DGS include the Gaussian primitives, the differentiable rasterizer, the training pipeline, and the compression and streaming mechanisms. Because a scene may contain millions of Gaussians, efficient compression is essential for storage and transmission over bandwidth-constrained networks. 3GPP is studying how to represent 3DGS in a standardized way, potentially using existing media codecs or new formats. The renderer must also handle sorting and blending of Gaussians, which is a performance-critical operation. In addition, 3DGS supports view-dependent colour through spherical harmonics, enabling realistic reflections and lighting.

The role of 3DGS in the network is to enable immersive, six-degrees-of-freedom (6DoF) experiences for XR and volumetric video services. Compared to traditional mesh-based representations, 3DGS offers higher visual fidelity and more efficient rendering, making it attractive for real-time applications such as telepresence, gaming, and virtual collaboration. 3GPP's study in Rel-19 aims to understand the requirements and potential standardization of 3DGS for 5G-Advanced and future networks, including how to deliver such content over 5G with low latency and high reliability.

Overall, 3DGS is an emerging media technology that 3GPP is evaluating for inclusion in its media services framework. Its study in TS 26.956 reflects the growing importance of immersive media and the need for standardized formats that can be efficiently delivered over cellular networks.

Purpose & Motivation

3D Gaussian Splatting exists to solve the problem of real-time rendering of photorealistic 3D scenes for immersive media. Traditional volumetric representations such as point clouds and meshes struggle to achieve both high visual fidelity and real-time performance, especially for complex scenes with view-dependent effects. Neural radiance fields (NeRF) produce high-quality results but require expensive neural network inference, making them unsuitable for real-time applications on mobile devices. 3DGS addresses this by using explicit Gaussian primitives that can be rendered efficiently with a rasterization pipeline.

Historically, 3DGS was introduced in the computer graphics community in 2023 as a breakthrough for real-time radiance field rendering. Its ability to train quickly and render at high frame rates made it immediately attractive for XR and gaming. 3GPP recognized its potential for immersive media services and initiated a study in Release 19 under TS 26.956 to investigate how 3DGS content can be represented, compressed, and delivered over 5G networks.

The motivation for 3GPP to study 3DGS is to ensure that future immersive services can be standardized and interoperable. Without a common format, content creators and service providers would face fragmentation, and network operators would lack guidance on bandwidth and latency requirements. By studying 3DGS, 3GPP aims to define media handling procedures that enable efficient streaming and rendering of Gaussian splatting content, thereby supporting the growth of XR and volumetric media in 5G-Advanced networks.