hash stringlengths 32 32 | doc_id stringlengths 5 12 | section stringlengths 5 1.47k | content stringlengths 0 6.67M |
|---|---|---|---|
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.12.1.4 Media Application Service model
| In order to define a media delivery architecture, it is assumed that an application, if following certain assumptions, can benefit from the media delivery architecture.
Terminology used in the common Media Delivery architecture as defined in TS 26.501 [4], clause 4.1.2, and TS 26.506 [5], clause 4.1.2, is used withou... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.12.1.5 Key Issues
| In order to define a Media Delivery system for a diverse set of media applications and services traffic patterns, the following are studied taking the 5G Media Delivery architecture as a starting point for discussion:
1. Should the media delivery architecture for streaming and real-time communication services be harmo... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.23 Work topic #2: 6G media
| Editor's note: The present Work Task is structured according to the agreed subsection format, including: Description, Key Issues, Context and External Factors, Potential Solutions, Mapping of Issues to Solutions, and Conclusions. The subsection ordering may be adapted as appropriate for the specific content of the Work... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.34 Work topic #3: Media aspects related to SA2 topics
| Editor's note: The present Work Task is structured according to the agreed subsection format, including: Description, Key Issues, Context and External Factors, Potential Solutions, Mapping of Issues to Solutions, and Conclusions. The subsection ordering may be adapted as appropriate for the specific content of the Work... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45 Work topic #4: Media for ubiquitous access
| Editor's note: The present Work Task is structured according to the agreed subsection format, including: Description, Key Issues, Context and External Factors, Potential Solutions, Mapping of Issues to Solutions, and Conclusions. The subsection ordering may be adapted as appropriate for the specific content of the Work... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45.1 Introduction
| |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45.1.1 High-level Description
| Editor's note: improved description needed.
This clause addresses the study aspects and opportunities for support of media services on ubiquitous networks including Non-Terrestrial Networks and other low bit-rate/low power scenarios beyond speech. The primary focus is to identify supported bitrates, functionalities, d... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45.1.2 Potentially relevant use cases and requirements
| Editor's note: needs to be completed by checking SA1.
|
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45.1.3 Potentially relevant 6G architecture key issues
| The following key issues in TR 23.801-01 may potentially be relevant for this work topic:
- Key Issue #4: User Plane Architecture
- Key Issue #23: Support of 6G NTN
- Key Issue #24: Analyse 5GS IoT features and solutions
|
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.45.1.4 Key Issues
| In order to identify aspects and opportunities for support of media services on ubiquitous networks including Non-Terrestrial Networks and other low bit-rate/low power scenarios, the following key issues are studied as a starting point for discussion:
1. What are bitrate ranges, latencies and loss characteristics of r... |
c2ab708c7c1b780cd97ba4af5f967fa5 | 26.870 | 6.56 Work topic #5: Trusted and private communication for media
| Editor's note: The present Work Task is structured according to the agreed subsection format, including: Description, Key Issues, Context and External Factors, Potential Solutions, Mapping of Issues to Solutions, and Conclusions. The subsection ordering may be adapted as appropriate for the specific content of the Work... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1 Scope
| This clause shall start on a new page.
The present document …
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 2 References
| The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
- References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
- For a specific reference, subsequent revisions do not apply.
-... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 3 Definitions of terms, symbols and abbreviations
| This clause and its three subclauses are mandatory. The contents shall be shown as "void" if the TS/TR does not define any terms, symbols, or abbreviations.
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 3.1 Terms
| For the purposes of the present document, the terms given in 3GPP TR 21.905 [1] and the following apply. A term defined in the present document takes precedence over the definition of the same term, if any, in 3GPP TR 21.905 [1].
3DGS tile: a spatial volume of the scene represented by a specific bounding volume, conta... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 3.2 Symbols
| For the purposes of the present document, the following symbols apply:
Symbol format (EW)
<symbol> <Explanation>
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 3.34 Abbreviations
| For the purposes of the present document, the abbreviations given in 3GPP TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in 3GPP TR 21.905 [1].
3DGS 3D Gaussian Splatting
SH Spherical Harmonic
PLY Polygon f... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 4 3DGS representation format
| [Editor’s note: Placeholder for the description of the 3DGS format and characteristics]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 4.1 Introduction
| A 3D Gaussian Splatting (3DGS) scene is represented as a set of continuous primitives, anisotropic 3D Gaussians, each carrying geometric parameters and radiometric attributes. It was first introduced in 2023 in the research paper 3D Gaussian Splatting for Real-Time Radiance Field Rendering from INRIA [aa]. The data mod... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 4.2 Primitives
| [Editor’s note: align the definition with what the industry develops]
A 3DGS primitive is an oriented 3D Gaussian with the following attributes. The items below describe data elements, independent of any specific encoding:
- Position: 3D scene position of the primitive expressed with x, y, and z coordinates in the l... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 4.3 Camera parameters
| To ensure accurate and high-quality rendering, it is important to reuse the position and settings of the cameras used to capture the 3DGS scenes during the rendering process.
For each acquired view, complete camera information may be necessary: extrinsic parameters (pose as a matrix or quaternion translation in the sc... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5 Use cases
| [Editor’s note: Placeholder for the description of the use cases]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.1 Introduction
| The present clause describes service scenarios illustrating the generation and the consumption of 3DGS scenes as well as associated working assumptions on the service configurations that serve as basis for detailed analysis documented in the following clauses of this technical report.
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.2 On-device capture and sharing of a static 3DGS scene
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.2.1 Description
| A user initiates a short capture session on a mobile device (UE) using the rear or front camera(s). Various typical capture patterns may be supported, per example:
- Object/person sweep ("object scan", "3D selfie"): the user moves around a subject at close range, recording multiple viewpoints to ensure good coverage a... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.2.2 Working assumptions
| This clause outlines the end-to-end processing chain, from capture to rendering on the receiving UE. It enumerates the key functional blocks, and device capability requirements.
- Acquisition and 3DGS content generation
- Sensor capture (RGB video, potentially depth and position and orientation if available).
- On-d... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.3 Exploration of a large 3DGS environment
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.3.1 Description
| In this scenario the user explores a large 3DGS environment on a UE with responsive 6DoF or constrained-6DoF navigation.
A user launches an application and selects a large 3DGS scene (e.g., museum, mall level, outdoor plaza, city, …). The UE requests visible parts of the scenes around the current pose and prefetches l... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.3.2 Working assumptions
| This clause outlines the end-to-end processing chain, emphasising adaptive delivery and device capability requirements.
- Acquisition and content generation
- The capture and the generation of large 3DGS scenes are not addressed in this use case.
- Based on the 3DGS models, the region-based parts of the 3DGS scenes ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.4 Dynamic 3DGS content
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.4.1 Description
| A UE receives time-varying 3DGS content depicting a dynamic subject or scene (e.g., a performer, dancer, singer, exhibition moment, band, sport action …). The UE renders the 3DGS content sequence in real time. The delivery and rendering process may also be assisted by the network through mechanisms such as partial deli... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.4.2 Working assumptions
| This clause outlines the end-to-end processing chain covering the delivery and rendering of dynamic 3DGS content, emphasising adaptive delivery and device capability requirements.
- Acquisition and content generation
- The capture and the generation of dynamics 3DGS models are not the focus of this use case, but the ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.5.1 Description
| A user is represented by an avatar that combines a rigged mesh and a 3D Gaussian Splat layer for fine detail. The sender transmits a time aligned animation stream and a static 3DGS as part of the base avatar. The animation stream drives the rig and blendshapes. The renderer composites mesh shading with 3DGS contributio... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.54.21 Working assumptions
| [Editor’s note: We expect refinement on the format later.]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6 Quality factors
| [Editor’s note: Placeholder for the description of the quality factors]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.1 Introduction
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.2 Discussion
| The objectives of FS_3DGS_MED cover a wide range of aspects in the end-to-end workflow related to 3DGS, including on those related to workflow aspects:
b. Consistent end-to-end quality across different capturing and rendering systems for 3DGS representations.
In order to be able to approach this objective it is neces... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.3 Complexity
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.3.1 Dynamic 3DGS
| Dynamic scene complexity may significantly impact the feasibility of dynamic 3DGS content on mobile platforms. High‑motion or structurally complex scenes tend to increase GPU memory usage, rendering load, bandwidth consumption, and thermal pressure on the device.
The following parameters can directly constrain achiev... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.4 Metrics
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.5 Data size
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.6 Mapping to 3GPP 5G QoS Identifier (5QI)
| This clause describes a mapping of a 3DGS service to standardized 3GPP 5G QoS Identifier (5QI) parameters specified in TS 23.501 [bo].
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.6.1 Background
| Below are some of the pre-defined 5QI values in TS 23.501 [bo] for services that have similar QoS characteristics as that of 3DGS. The table 6.6.1-1 presents a subset of table 5.7.4-1 of TS 23.501.
Table 6.6.1-1: 5QI of standardized services potentially similar to 3DGS.
5QI
Value
Resource Type
Default Priority L... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 6.6.2 3DGS Application flows for 5QI mapping
| The different use cases described in clause 5 of the present document focuses on delivery of number of application flows. These application flows include:
- Delivery of 3DGS application flows
- Delivery from/to the UE of static 3DGS scene content
- Delivery from the network to the UE of dynamic or view based 3DGS co... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7 Static 3DGS content creation
| [Editor’s note: Placeholder for the description of the 3DGS content creation processes]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7.1 3DGS generic workflow
| As an example, a 3DGS model construction from the capture of 2D data follows the production workflow as illustrated in figure 7.1-1:
Figure 7.1-1: 3DGS model production workflow from a 2D video
The workflow consists of three parts. First, the capture phase. During this phase, numerous views of a real-world object ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7.2 Capture
| Content acquisition for 3D Gaussian Splats (3DGS) relies on capturing accurate 3D data from real-world objects and environments. Primary methods include:
- 2D image and video capture: video sequences or sets of 2D images capture from various positions and orientations offer coverage of dynamic or complex environments.... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7.3 Structure from Motion
| The Structure from Motion (SfM) step consists in operating the feature extraction and matching and retrieving the camera parameters when unknown. After image alignment, the process creates a sparse point cloud that is further densified with depth calculation methods.
Camera parameters, often known by the capturing sys... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7.4 Training
| 7.4.1 Introduction
The third step is about the creation of the gaussian splats associated to each 3D point using iterative optimization process that will search for the splats that match as much as possible the source video for a given pose (position + orientation) by optimizing the size, shape, colour and transparen... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 7.4.5 Discussion
| The original paper on 3D Gaussian Splatting for Real-Time Radiance Field Rendering [aa] published in August of 2023 presents a workflow as shown in figure 7.4.5-1 below.
Figure 7.4.5-1: Organization of the workflow [aa]
Whilst the original workflow process includes a closed loop optimization problem using converge... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 8 3DGS encapsulation and delivery formats
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 8.1 Introduction
| As a representation format for 3D media, 3D Gaussians support large scalability in terms of the real-size volume of the scene represented, as well as the number of 3D Gaussians which can be used to represent the scene or object. For this reason, it is important that a 3DGS encapsulation and/or delivery format supports ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 8.2 Requirements
| To support the efficient data access and delivery of 3DGS data, the following encapsulation and delivery requirements related to 1) the position based random access of 3D Gaussians and 2) the delivery and/or rendering of different levels of detail (LODs) for the scene, are identified:
1. A method to identify 3DGS data... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 8.3 3DGS tiles
| One method to associate spatial volumes with 3D Gaussians in a scene is by defining 3DGS tiles.
A 3DGS tile is a spatial volume of the scene represented by a specific bounding volume, containing a set of 3D Gaussians for a given level of detail (LOD).
[Editor’s Note: More details needed.]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 8.4 Compression aspects for investigation
| Encapsulation and delivery formats are inherently related to the technologies used to compress 3DGS representation data. To ensure the support of 3DGS spatial random access and LOD for rendering and delivery, the following should be investigated regarding 3DGS compression technologies:
1. The ability to support a cert... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 98 3DGS rendering
| [Editor’s note: Placeholder for the description of the 3DGS rendering processes]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 98.1 Pipeline description
| The commonly used pipeline to render 3DGS objects is outlined in this clause. It defines the per-frame inputs, configurable options, and outputs of a representative implementation, without prescribing specific algorithms.
The inputs are, per 3DGS frame and per rasterized image:
- 3DGS data: per-Gaussian attributes ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 98.2 Rasterization process
| 98.2.1 Introduction
This section clause presents a rendering process used to render a 3DGS model. A 3DGS model is rendered based on the observer’s position and orientation and on the Gaussian, primitives defined in Section clause 4.1. Depending on the chosen representation format, the rasterization described below m... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 109 High level media data workflows
| [Editor’s note: Placeholder for the description of the workflows]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 109.1 All-in-client configuration
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.1.1 Description
| This section outlines different aspects of the media data workflows, of the use cases described in clause 5 of the present document, that primarily corresponds to the functionality that may run on the UE.
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.1.2 Media workflow steps
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.1.2.1 Workflow description
| Following are some of the media workflow steps of different use cases described in clause 5 of the present document that can be run on the UE using an All-in-client configuration:
- Scene capture on the UE, for example using the rear or front camera(s) of a mobile device. Multiple viewpoints may be captured to ensure ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.1.2.2 Characteristics
| Below are the characteristics for the all-in-client configuration for realizing the use case:
- Latency/Performance: Dependent upon the capabilities of the UE device that is capturing, generating, or rendering the 3DGS asset
- Scalability: Limited by the UE device capabilities (local memory, GPU memory, decoding capa... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.1 Description
| This section outlines different aspects of the media data workflows, of the use cases described in clause 5 of the present document, that primarily corresponds to the functionality that may be split between the client and the server (network). This configuration supports interactive navigation in large or dynamic 3DGS ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.2 Media Workflow Steps
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.2.1 Workflow description
| Following are some of the media workflow steps of different use cases described in clause 5 of the present document that run in the Client-Server configuration (e.g., an edge network application server).
- Generation of 3DGS content using 2D captures of the scene (clause 5.2)
- Generation of dynamic 3DGS content and/... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.2.2 Characteristics
| Below are the characteristics for the client-server configuration for realizing the use cases described in clause 5 of the present document:
- Latency/Performance: Dependent upon network and application latency based on:
- the capabilities of the network server that is generating the 3DGS content
- UE device that ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.3 Workflow with capability negotiation
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.3.1 Overview
| In the client-server configuration, the heavy lifting of content preparation and storage is handled by the server, while the UE acts as the rendering endpoint. Given the significant variability in 3DGS scene complexity (ranging from thousands to millions of primitives) and the diverse performance profiles of mobile UEs... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.3.2 Objectives
| To ensure consistent Quality of Experience (QoE), specifically regarding frame rate stability and thermal management, the delivery session is initiated with a capability exchange phase. This clause describes a delivery workflow where the content is optimized based on a negotiation of device capabilities. To support dif... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.3.3 Workflow
| To ensure consistent Quality of Experience (QoE) and prevent device overheating, the delivery session follows a negotiated workflow. This process begins with a session initialization phase where the UE receives Service Access Information (SAI), containing essential bootstrapping parameters (e.g., Provisioning Session I... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.4.1 Overview
| For large-scale environments, such as city-scale digital twins (addressing the use case in clause 5.4), the entire scene typically exceeds the storage and rendering capacity of the UE. In this context, the delivery workflow must be extended to incorporate the user's spatial context dynamically.
In addition to the init... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.4.2 Spatial optimization strategies
| The environment may be partitioned into spatial 3DGS tiles, with each tile defined at different levels of detail (LOD). The server selects the appropriate tiles and their required LOD based on its proximity and visibility within the user's frustum. High-density tiles (e.g. LOD 4) are assigned to the center of the viewp... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 10.2.4.3 Workflow
| During initialization, the server determines a rendering budget based on the UE's hardware report. This budget remains valid for the session (unless capabilities change). Subsequently, during the delivery phase, the server uses this pre-determined budget to filter and adapt the large-scale scene based on the received s... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1011 Mapping to the 3GPP services
| [Editor’s note: Placeholder for the description of the 3GPP services used]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1011.1 All in UE configuration
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1110.2 Client-server configuration
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1112 Related activities and products and services
| [Editor’s note: Placeholder for the description of the products and services]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1211.1 Standardization activities
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1211.2 Services
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1211.3 Software and products
|
With the increase in popularity of 3DGS, software and products related to 3DGS generation and rendering have proliferated. Some of these tools started out as 3D scanning or photogrammetry tools, but have added support for 3DGS generation, rendering, post-processing, etc. Below some popular consumer software and produ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1312.1 Capture
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 1312.2 Transmission
| 1312.34 Rendering
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.1 Mobile renderer features
| A 3DGS player on mobile platforms aims at validating the rendering pipeline on mobile devices. The application may be designed to be built using hybrid architecture:
- Native layer (e.g. C++): Handles the core rendering tasks using OpenGL ES 3.2. It implements a tile-based rasterizer inspired by the original 3DGS meth... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.2 Rendering process on mobile platforms
| The rendering process for 3DGS relies on a hybrid CPU/GPU architecture optimized for the rasterization of semi-transparent volumes. Unlike classic mesh rendering which utilizes the Z-buffer for occlusion management, 3DGS requires strict alpha blending, necessitating that primitives be drawn from farthest to nearest (ba... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.3 Benchmark methodology
| The renderer includes a benchmarking mode that allows modifying rendering parameters dynamically. This mode may use the thermal management API to maintain consistent clock speeds during benchmarking. In this mode, the AR runtime processes and the AR environment are disabled to ensure a fair comparison and a frame per s... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.4 Preliminary experimental results
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.4.1 Test conditions
| An evaluation of a mobile rendering capabilities was conducted with the following parameters
• Device: Google Pixel 9a (Tensor G4 chipset, middle-range device, March 2025).
• Application : Tencent 3DGS mobile player
• Build configuration: Release mode with optimizations enabled.
• Test duration: 30 ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.4.2 Impact of the number of points
| The following table 13.3.4.2-1 and figure 13.3.4.2-1 illustrate the performance impact when rendering the same scene with an increasing number of Gaussian primitives (SH degree fixed at 3).
Table 13.3.4.2-1: Performance for various numbers of points (device=Pixel 9a, spherical harmonics degree=3)
Points
Frame per se... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.4.3 Impact of spherical harmonics degree
| The following table 13.3.4.3-1 and figure 13.3.4.3-1 analyse the cost of higher-order color view-dependence processing on the full model (~485k points).
Table 13.3.4.3-1: Performance for various spherical harmonics degrees (device=Pixel 9a, points=485k)
SH Degree
Frame per second
CPU load (%)
GPU load (%)
Ets. P... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 13.3.4.4 Analysis
| These results suggest that allocating resources for the rendering stage of static 3DGS scenes is feasible on current-generation mobile hardware for managed complexities (e.g., < 200k visible points), though this must be balanced against the requirements of the full delivery pipeline.
Variations were observed between t... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 3.3 Abbreviations
| For the purposes of the present document, the abbreviations given in 3GPP TR 21.905 [1] and the following apply. An abbreviation defined in the present document takes precedence over the definition of the same abbreviation, if any, in 3GPP TR 21.905 [1].
3DGS 3D Gaussian Splatting
SH Spherical Harmonic
PLY Polygon f... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 5.5.2 Working assumptions
| [Editor’s note: We expect refinement on the format later.]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9 3DGS rendering
| [Editor’s note: Placeholder for the description of the 3DGS rendering processes]
|
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9.1 Pipeline description
| The commonly used pipeline to render 3DGS objects is outlined in this clause. It defines the per-frame inputs, configurable options, and outputs of a representative implementation, without prescribing specific algorithms.
The inputs are, per 3DGS frame and per rasterized image:
- 3DGS data: per-Gaussian attributes ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9.2 Rasterization process
| |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9.2.1 Introduction
| This clause presents a rendering process used to render a 3DGS model. A 3DGS model is rendered based on the observer’s position and orientation and on the Gaussian, primitives defined in clause 4.1. Depending on the chosen representation format, the rasterization described below may need to be updated to consider the f... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9.2.2 Main stages
| The main steps of the rasterization process are:
- Culling: all Gaussians are examined, and those that do not affect the image are eliminated: elements located behind the observer or entirely outside the targeted display area are removed.
- Sorting: the remaining Gaussians are then sorted by distance from the observ... |
de4351bc6585aa9d49d11017627c9c5f | 26.958 | 9.2.3 Detailed implementation
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.