3D Mesh to Point Cloud Conversion for Gaussian Splat Rendering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing 3D modeling techniques, such as 3D Gaussian splatting, are not compatible with polygonal meshes, limiting the use of existing 3D models and tools designed for meshes in radiance field rendering.

Innovation Solution

A method is developed to generate a point cloud representation of a 3D model by performing ray tracing from a camera view, determining intersection locations, and using 3D Gaussian functions to define the extent and color of points in the point cloud, allowing integration with existing 3D models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If 3D Gaussian splatting is used for radiance field rendering, then rendering quality and efficiency are improved, but compatibility with existing polygonal mesh tools and models is lost

Engineering Contradiction:
Improverendering efficiencyVSAvoidcompatibility with existing 3D tools
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent introduces a conversion process that acts as an intermediary between polygonal meshes and 3D Gaussian splatting. The system takes existing polygonal mesh models as input, automatically converts them into point cloud representations with Gaussian splats, and outputs formats compatible with 3DGS rendering pipelines. This intermediary conversion layer preserves the benefits of 3DGS while maintaining compatibility with the vast library of existing mesh-based 3D assets and tools.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of manufacture

If polygonal meshes are used for 3D modeling, then ease of design and creation is maintained, but integration with radiance field approaches is not possible

Engineering Contradiction:
Improveease of 3D model creationVSAvoidintegration with radiance fields
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent segments the polygonal mesh model into discrete points that can be represented as Gaussian splats. By dividing the continuous mesh surface into a point cloud representation, the system enables each point to be independently processed as a Gaussian function, thereby bridging the gap between traditional mesh geometry and radiance field requirements while preserving the original mesh creation workflow.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If existing 3D models are converted to point clouds manually, then accuracy can be maintained, but time consumption and complexity increase

Engineering Contradiction:
Improvepoint cloud accuracyVSAvoidconversion time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements preliminary automated processing that performs the mesh-to-point-cloud conversion in advance, before rendering. The system pre-processes polygonal meshes by generating optimized point cloud representations with appropriate Gaussian parameters, storing these converted formats for later use. This preliminary conversion action eliminates the need for manual processing during actual rendering workflows, maintaining high accuracy while significantly reducing time consumption.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4723048A1Methods and apparatuses for 3D rendering
Publication Date: 2026.04.08 SONY INTERACTIVE ENTERTAINMENT LLC
  • EP4723048A1 patent drawingFigure 1
  • EP4723048A1 patent drawingFigure 2
  • EP4723048A1 patent drawingFigure 3

AI summary

There is provided a computer-implemented method of generating a point cloud, the method comprising: receiving first data representing a 3D model, the 3D model comprising a plurality of surfaces; performing ray tracing from a camera view, wherein the ray tracing comprises defining one or more rays passing through a location of the camera view; determining one or more intersection locations, corresponding to the location of an intersection of a ray with one or more surfaces of the 3D model; and generating a point cloud representation of the 3D model from the one or more intersection locations.