Adaptive 3D Gaussian Coefficients for Efficient Scene Rendering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The high memory consumption and computational complexity of 3D Gaussian (3DG) splatting for 3D scene rendering, particularly in consumer devices and low-end GPUs, due to the large number of coefficients required for spherical harmonics, especially for specular surfaces, hinder its deployment and efficiency.

Innovation Solution

Adaptive computation of the number of coefficients for each 3D Gaussian or group of 3DGs based on the scene's material properties, using lower degree harmonics for Lambertian surfaces and higher degree harmonics for specular surfaces, reducing memory footprint and computational load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If higher degree spherical harmonics are used for all surfaces, then rendering quality for specular surfaces is improved, but memory consumption and computational complexity increase

Engineering Contradiction:
Improverendering qualityVSAvoidmemory consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies different degrees of spherical harmonics to different surfaces based on their material properties. Specular surfaces use higher degree harmonics (e.g., degree 4 or 8) to capture accurate reflections, while Lambertian surfaces use lower degree harmonics (e.g., degree 0-2) to reduce computational load. This local differentiation resolves the contradiction by allocating computational resources only where needed for high rendering quality.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts the degree parameter of spherical harmonics based on surface material classification. By changing this parameter adaptively rather than using a fixed high degree for all surfaces, the system maintains high rendering quality for specular surfaces while significantly reducing memory consumption and computational complexity for Lambertian surfaces.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If higher degree spherical harmonics are used for all surfaces, then rendering quality for specular surfaces is improved, but computational complexity increases

Engineering Contradiction:
Improverendering qualityVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies different degrees of spherical harmonics to different surfaces based on their material properties. Specular surfaces use higher degree harmonics (e.g., degree 4 or 8) to capture accurate reflections, while Lambertian surfaces use lower degree harmonics (e.g., degree 0-2) to reduce computational load. This local differentiation resolves the contradiction by allocating computational resources only where needed for high rendering quality.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts the degree parameter of spherical harmonics based on surface material classification. By changing this parameter adaptively rather than using a fixed high degree for all surfaces, the system maintains high rendering quality for specular surfaces while significantly reducing memory consumption and computational complexity for Lambertian surfaces.

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If uniform number of coefficients is used for all Gaussians, then implementation simplicity is maintained, but memory requirements increase unnecessarily

Engineering Contradiction:
Improveimplementation simplicityVSAvoidmemory requirements
Core Design Contradiction:
Ease of manufactureVSQuantity of substance

Solution Approach 1:

The patent applies different degrees of spherical harmonics to different surfaces based on their material properties. Specular surfaces use higher degree harmonics (e.g., degree 4 or 8) to capture accurate reflections, while Lambertian surfaces use lower degree harmonics (e.g., degree 0-2) to reduce computational load. This local differentiation resolves the contradiction by allocating computational resources only where needed for high rendering quality.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent segments the scene into different material types (specular and Lambertian surfaces) and applies different coefficient configurations to each segment. This segmentation allows the system to use fewer coefficients for Lambertian surfaces while maintaining sufficient accuracy, thereby reducing overall memory requirements without significantly complicating the implementation.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4648427A1Adaptive attributes for 3D gaussians
Publication Date: 2025.11.12 INTERDIGITAL CE PATENT HOLDINGS SAS
  • EP4648427A1 patent drawingFigure 1
  • EP4648427A1 patent drawingFigure 2
  • EP4648427A1 patent drawingFigure 3

AI summary

Methods and apparatus are provided to implement adaptive attributes for three-dimensional Gaussians for image processing. The embodiments enable image rendering, encoding, decoding and other processes to be performed using the three-dimensional Gaussians. In at least one embodiment, the number of coefficients for each Gaussian or each group of Gaussians is adapted to a scene. In other embodiments, the number of parameters for each three-dimensional Gaussian or each group of three-dimensional Gaussians is stored and/or transmitted.