A method for generating a three-dimensional model guided by a multi-level regularized discrete code tree

By using a multi-level regularized discrete code tree method, the 3D model is decomposed into rings, contours, and solid parts. The code tree is generated using vector quantization adversarial learning and Transformer networks, which solves the problems of insufficient diversity and poor controllability in the generation of 3D models in the existing technology, and realizes efficient, diverse and controllable 3D model generation.

CN119848961BActive Publication Date: 2026-04-07XIAMEN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-12-31
Publication Date
2026-04-07

AI Technical Summary

Technical Problem

Existing 3D model generation methods are insufficient in terms of generation diversity, controllability, and coherence, resulting in scattered, inconsistent, and redundant generation results that fail to meet users' personalized needs.

Method used

A multi-level regularized discrete code tree method is adopted to divide the 3D model construction sequence into loop, contour and solid parts. Regularized discrete codebook is extracted through vector quantization adversarial learning, and code tree is generated by training a standard Transformer network. The 3D model is generated by combining sketching and extrusion methods.

Benefits of technology

It enables the generation of more realistic and diverse 3D models, enhances the controllability and efficiency of the generation process, reduces computational costs, and produces models with complete structures, rich details, and high visual quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119848961B_ABST
    Figure CN119848961B_ABST
Patent Text Reader

Abstract

A three-dimensional model generation method guided by multi-level regularized discrete code tree, relates to computer-aided design. The construction sequence corresponding to the three-dimensional model is divided into three parts of ring, contour and entity. The corresponding regularized discrete codebook is extracted by extracting the geometric information of the three parts respectively. The three-dimensional model construction sequence is represented as a regularized code tree containing three levels to obtain the compressed three-dimensional model construction sequence representation. The code tree is trained by a standard Transformer network. Based on the code tree as global information, the training condition guiding network is generated by sketch and extrusion method to generate the construction sequence corresponding to the three-dimensional model corresponding to the code tree. The generation of the three-dimensional model is guided by the multi-level regularized code tree, which significantly improves the quality, realism, diversity and complexity of the generated model. The generation effect is obviously improved compared with the existing highest baseline model.
Need to check novelty before this filing date? Find Prior Art