Three-dimensional model generation method, device, equipment, storage medium and program product

By representing the surface mesh with a binary signed distance field, optimizing vertex positions, and eliminating irrelevant voxels, a watertight 3D model is generated. This solves the problems of resource waste and generation errors in existing technologies and achieves efficient and high-precision 3D model generation.

CN122115785APending Publication Date: 2026-05-29BEIJING ZITIAO NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING ZITIAO NETWORK TECH CO LTD
Filing Date
2026-02-06
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing 3D model generation methods struggle to improve generation quality when training resources are controllable, and non-watertight models lead to wasted computational resources and generation errors.

Method used

The surface mesh is represented by a binary signed distance field. Only the vertex positions are optimized and irrelevant voxels are removed. The watertight model is reconstructed through a sparse mesh. High-quality training data within the observation range is used to reduce the computational resource requirements.

Benefits of technology

It improves the accuracy and efficiency of 3D model generation, reduces computational resource consumption, and generates models with correct and stable topology, suitable for generating 3D models with different observation ranges.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115785A_ABST
    Figure CN122115785A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of computers and discloses a three-dimensional model generation method, device, equipment, storage medium and program product, which comprises the following steps: performing three-dimensional reconstruction on a first object based on first information to obtain first surface voxels, wherein the first information is used for representing the characteristics of the first object; inputting the first surface voxels into a first model to generate a first three-dimensional model; wherein the first model is configured to be obtained by training a second model based on sample data; the sample data comprises second surface voxels of a third three-dimensional model in each observation range and a second three-dimensional model, the third three-dimensional model corresponds to a second object; the three-dimensional models of all observation ranges cover the third three-dimensional model; the third three-dimensional model is obtained by adjusting a vertex position, the vertex position corresponds to a surface mesh of a fourth three-dimensional model, the fourth three-dimensional model corresponds to the second object, and the surface mesh of the fourth three-dimensional model is represented by a binary signed distance field, so that the generation problem of the three-dimensional model can be solved.
Need to check novelty before this filing date? Find Prior Art