Interoperable compression for 3D Gaussian Splat representations
Gaussian Splats (often referred to as 3D Gaussian Splatting) have emerged as a practical representation for rendering realistic 3D objects and scenes using a set of 3D Gaussian primitives and associated attributes. MPEG is advancing Gaussian Splat Coding (GSC) through a two-layer strategy: a short-term path enabling interoperable coding within established point cloud compression frameworks, and a joint exploration path preparing future versions with higher coding efficiency and broader functionality.
Gaussian Splats in the MPEG context
Representation perspective
Gaussian Splats can be seen as a point-based 3D representation in which each primitive carries geometry-related parameters and rendering-oriented attributes. In MPEG terms, the representation is explicit (a set of coded parameters) and is therefore naturally connected to MPEG’s point cloud compression ecosystem. The optimization of the representation itself (e.g., deciding how many splats to use, or how splat density is distributed in space) is considered an application-side choice, while MPEG coding focuses on efficient, interoperable compression of the chosen representation.
Explicit parametersRendering-oriented attributesInteroperable codingStandards perspective
MPEG maintains two complementary standards for point cloud compression that are directly relevant to Gaussian Splats: V-PCC (video-based) and G-PCC (geometry-based). Gaussian Splats exhibit a wide range of characteristics depending on acquisition, content type, and authoring choices, which motivates providing short-term solutions in both frameworks, while jointly exploring new tools and representations for future versions.
V-PCCG-PCCProfiles & amendmentsCurrent standardization activities
MPEG’s short-term standardization path for GSC is based on defining dedicated profiles and extensions within existing standards. This approach enables rapid interoperability by leveraging mature toolsets, while keeping the door open to longer-term innovation through the joint exploration activities.
V-PCC-based Gaussian Splat Coding
A V-PCC-based solution is being developed through an amendment defining a Gaussian Splat profile within the existing V-PCC framework. The work leverages the video-based architecture of V-PCC and its patch/atlas coding concepts to map Gaussian Splat parameters into 2D representations suitable for compression with video coding technology. The amendment reached the Committee Draft Amendment (CDAM) stage at MPEG meeting 153.
Standard: ISO/IEC 23090-5 (MPEG-I V-PCC) — Amendment for Gaussian Splats (profile-based).
Video-basedProfile / amendmentCDAMG-PCC-based Gaussian Splat Coding
A complementary G-PCC-based solution is being developed through a dedicated profile and extensions within the G-PCC framework. This path focuses on efficient coding of explicit Gaussian parameters using geometry-based point cloud compression tools, while aligning with the existing G-PCC ecosystem and conformance methodology. The work advanced to a Working Draft (WD) stage at MPEG meeting 153.
Standard: ISO/IEC 23090-9 (MPEG-I G-PCC) — Profile / amendment work for Gaussian Splats.
Geometry-basedProfile / amendmentWDTwo-phase direction
The near-term focus is to deliver interoperable coding solutions for static Gaussian Splats through profiles and extensions within V-PCC and G-PCC. In parallel, the joint exploration prepares the next versions of Gaussian Splat Coding by studying improved efficiency and broader capabilities, including dynamic content and advanced toolsets.
Joint exploration on Gaussian Splat Coding
In addition to the short-term profile-based work, MPEG is conducting a joint exploratory activity on Gaussian Splat Coding. The exploration aims to identify and evaluate new representation and coding technologies and to prepare the ground for future standardization steps, including potentially broader calls for evidence or proposals.
Core technology
The exploration investigates how Gaussian parameters and attributes can be coded more efficiently, including aspects such as quantization, parameter ordering, prediction, and attribute handling. The goal is to enable fair comparison of coding tools under common test conditions and to converge toward robust technical solutions.
Tools explorationEfficiencyStatic & dynamic content
While short-term amendments focus primarily on lightweight intra coding for static content, the exploration considers longer-term needs including dynamic Gaussian Splats, temporal consistency, and improved rate–distortion performance.
Dynamic GSTemporal aspectsSoftware consolidation
A consistent evaluation environment requires aligned software tools, reproducible pipelines, and cross-checkable results. The exploration therefore includes consolidation of tooling used for processing, encoding, decoding, and rendering-based evaluation.
ReproducibilityCross-checkEvaluation framework
Gaussian Splats are evaluated using an approach consistent with MPEG practice: common test conditions, stable anchors, and reproducible pipelines. Objective metrics are computed on rendered images under fixed rendering conditions and viewpoint sets, enabling fair and comparable measurements of coding-induced distortion. Subjective evaluation, when conducted, requires careful selection of sequences and pose traces to avoid bias from source-generation artefacts and to ensure that observers score compression artefacts rather than unrelated reconstruction issues.
Anchors and common test conditions
Anchors provide stable reference points for comparison and are expected to rely on mature, reproducible implementations. In the GSC context, anchors based on existing MPEG standards (including V-PCC and G-PCC profiles) provide a common baseline across contributions.
AnchorsCTCObjective & subjective evaluation
Objective evaluation focuses on rendered-view metrics and controlled pipelines. Subjective evaluation is sensitive to content and viewpoint selection; pose traces are designed to stay within valid capture regions and avoid pre-existing artefacts that would otherwise dominate human scoring.
Rendered metricsPose tracesRelationship with other MPEG standards and inter-WGs collaboration
GSC is being progressed through coordinated activities across MPEG groups, reflecting the fact that Gaussian Splats intersect point cloud coding, video coding technology, and system-level carriage considerations. The short-term standardization work is carried out within the established V-PCC and G-PCC frameworks, while the joint exploratory activity supports broader technical alignment and longer-term evolution. Inter-WG collaboration helps ensure consistent technical assumptions, comparable evaluation conditions, and convergence toward interoperable solutions.
| Area | How it is addressed for GSC |
|---|---|
| Point cloud coding frameworks | Profile and amendment work within V-PCC (video-based) and G-PCC (geometry-based), maintaining continuity with existing MPEG-I standards. |
| Exploration and alignment | Joint exploration involving WG 7, WG 4, and JVET to study new coding tools, representations, evaluation methodology, and longer-term directions such as dynamic GS. |
| Carriage and encapsulation | Studied in coordination across groups, with decisions taken within the appropriate standardization frameworks (e.g., V3C/V-PCC-based integration paths and other approaches under consideration). |
Resources
Public reference software (related standards)
MPEG’s open-source PCC reference software is available publicly and can serve as a practical starting point to understand the baseline technologies used for profile-based GSC work.
V-PCC Test Model (TMC2): github.com/MPEGGroup/mpeg-pcc-tmc2
G-PCC Test Model (TMC13): github.com/MPEGGroup/mpeg-pcc-tmc13
Renderer and visualization
Practical evaluation of Gaussian Splats frequently relies on rendered-view analysis. MPEG also maintains open-source rendering tools for 3D point clouds and meshes that are useful for visualization and experimentation in related MPEG-I workflows.
MPEG 3DG renderer: github.com/MPEGGroup/mpeg-3dg-renderer
VisualizationRenderingHow to follow the work
This page summarizes the publicly visible high-level status of MPEG GSC activities. Technical details, evaluation results, and milestone documents are progressed through MPEG meetings and associated exploration activities.