Polygonal Mesh Vertex Removal for Reducing Data Redundancy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing polygonal mesh generation techniques for computer graphics result in high data redundancy, leading to significant processing overhead and poor performance in real-time applications, despite efforts to optimize using machine learning algorithms.

Innovation Solution

A method and system for generating and adapting polygonal mesh data by comparing vertex counts and quality parameters between reference and modified meshes, using machine learning to determine candidate functions for vertex removal, thereby reducing redundancy while maintaining visual quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If surface subdivision algorithms are used to generate polygonal meshes, then the mesh can represent the surface geometry, but the mesh structure contains large numbers of vertices, edges and faces resulting in high data redundancy

Engineering Contradiction:
Improvesurface representation accuracyVSAvoidnumber of vertices, edges and faces
Core Design Contradiction:
Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and removes redundant vertices from the polygonal mesh while preserving the essential surface geometry. The system identifies and eliminates duplicate or unnecessary vertices that do not contribute to the visual quality of the rendered object, thereby reducing data redundancy while maintaining manufacturing precision.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of starting with a low-poly mesh and adding detail, the patent inverts the traditional approach by starting with a high-poly mesh generated by surface subdivision algorithms and systematically removing redundant elements. This inversion allows the system to begin with accurate surface representation and progressively simplify the mesh structure.

Inventive Principle:
Principle #13The other way round (Inversion)

2Manufacturing precision

If high vertex count meshes are used to maintain visual quality, then the surface representation is accurate, but processing overhead increases significantly

Engineering Contradiction:
Improvevisual qualityVSAvoidprocessing efficiency
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The patent applies local quality optimization by selectively removing vertices based on their individual contribution to visual quality. The system evaluates each vertex's importance to the surface geometry and removes only those vertices that do not significantly impact visual appearance, thereby maintaining overall visual quality while improving processing efficiency.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the parameter of vertex count from high to optimized-low while maintaining visual quality through intelligent selection criteria. The system dynamically adjusts which vertices to remove based on geometric analysis, ensuring that visual quality parameters are preserved even as the total vertex count decreases.

Inventive Principle:
Principle #35Parameter changes

3Quantity of substance

If machine learning algorithms are applied to optimize mesh, then data redundancy can be reduced, but the system complexity increases

Engineering Contradiction:
Improvedata redundancyVSAvoidsystem complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent implements self-service by training the machine learning model to autonomously identify and remove redundant vertices without requiring complex external optimization systems. The trained model independently analyzes mesh structures and makes decisions about vertex removal, reducing the need for additional complex control systems.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent applies preliminary action by pre-training the machine learning model on a dataset of meshes before deployment. This preliminary training phase allows the system to learn optimal vertex removal strategies in advance, so that during actual mesh processing, the system can efficiently reduce data redundancy without requiring complex real-time decision-making infrastructure.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3736776B1Apparatus, system and method for the generation of polygonal meshes
Publication Date: 2023.10.25 SONY INTERACTIVE ENTERTAINMENT LLC
  • EP3736776B1 patent drawingFigure 1a~1b
  • EP3736776B1 patent drawingFigure 2
  • EP3736776B1 patent drawingFigure 3a~4b

AI summary

A method of generating a data set for training a system to generate data of a polygonal mesh by adapting data of an input polygonal mesh comprises obtaining data of a plurality of reference polygonal meshes for use in rendering a content, each reference polygonal mesh comprising a plurality of vertices for representing a surface in the content and for each reference polygonal mesh: determining a vertex count of the reference polygonal mesh, generating a second polygonal mesh by adding a predetermined number of vertices to the reference polygonal mesh, comparing a representation of the second polygonal mesh with a representation of the reference polygonal mesh, assigning a first value of a quality parameter to the reference polygonal mesh, and assigning a second value of the quality parameter to the second polygonal mesh based on the comparison of the representations, the value of the quality parameter for a given polygonal mesh indicative of a degree of difference between a representation of the given polygonal mesh and a representation of the reference polygonal mesh, and generating a data set comprising data indicative of the vertex count for the reference and second polygonal meshes and the value of the quality parameters for the reference and second polygonal meshes.