D-FRAME: Direction-Field-Based Wireframe Extraction for Complex CAD Models

Oct 26, 2026

Extracting wireframes from CAD point-cloud data remains challenging due to point-cloud defects (lack of orientation, noise, and sparsity) and the high density of closely spaced sharp edges in complex geometries. This work introduces D-FRAME, a multi-stage wireframe extraction framework that uses a direction field to improve edge detection quality and connectivity, refines sparse or noisy edge points, and performs a coarse-to-fine connection step to obtain robust wireframes. Combining restricted Voronoi diagram integration yields faithful CAD reconstruction, with strong results on synthetic and real scanned datasets.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Introduction

Related Work

Overview of D-FRAME

Feature Extraction and Prediction Branches

Edge Refinement Module

Direction-Field-Guided Wireframe Extraction

D-FRAME: Direction-Field-Based Wireframe Extraction for Complex CAD Models Yuan Feng , Honghao Dai , Guangshun Wei , Long Ma , Pengfei Wang, Yuanfeng Zhou , and Ying He , Member, IEEE Abstract—Extractingwireframes from CAD models represented by point cloud remains a signiﬁcant challenge in computer graphics. This difﬁculty arises from two main factors: ﬁrst, imperfections in the point cloud data, such as lack of orientation, noise, and sparsity; and second, the inherent complexity of geometric shapes, which often feature a high density of sharp edges in close proximity. In this paper, we propose D-FRAME, a multi-stage wireframe extraction framework that incorporates a novel direction ﬁeld to improve edge detection quality and connectivity, areﬁnement strategy to address sparse or noisy edge points, and a ﬁnal coarse-to-ﬁne connection module to extract a robust wireframe. The direction ﬁeld not only facilitates connectivity but also enhances the precision of extracted edges by mitigating the impact of misclassiﬁed points. By combining the Restricted Voronoi Diagram (RVD) with the extracted wireframes and the original point cloud, our approach also achieves highly faithful reconstruction ofCAD model. Experiments conducted on synthetic and real-world scanned CAD datasets demonstrate that D-FRAME effectively manages noise, sparsity, and complex geometries, yielding high-ﬁdelity wireframes. Index Terms—CAD modeling, direction ﬁelds, edge point classiﬁcation, wireframe extraction. I. INTRODUCTION THE extraction of wireframes from CAD-type point clouds is a fundamental and critical challenge in computer graphics and geometric modeling. Accurate wireframes are essential for effectively reconstructing CAD models with sharp features. However, achieving precise wireframe extraction from point clouds is hinderedby two primary challenges:(1)defects inpoint cloud data, such as a lack of orientation, noise, and sparsity, and Received11April2025;revised10August2025;accepted8September2025 . Date of publication 12 September 2025; date of current version 10 November 2025. This work was supported in part by the National Natural Science Foundation of China under Grant 62172257, Grant 62572284, 62502273, and Grant 62402295, in part by the Natural Science Foundation of Shandong Province (Major Basic Research) under Grant ZR2024ZD12, Grant ZR2025QC1558, and Grant ZR2024QF087, in part by the Shandong Key Laboratory of Digital and Data Convergence Publishing Media, and in part by the Ministry of Education, Singapore, under its Academic Research Fund under Grant RT19/22 . Recommended for acceptance by X. Guo. (Corresponding authors: Guangshun Wei; Yuanfeng Zhou.) Yuan Feng, Honghao Dai, Guangshun Wei, Long Ma, Pengfei Wang, and Yuanfeng Zhou are with the School of Software, Shandong University, Jinan 250100, China ([e-mail: yfeng@mail.sdu.edu.cn](e-mail: yfeng@mail.sdu.edu.cn); [dhhtang@163.com](dhhtang@163.com); [guangshunwei@gmail.com](guangshunwei@gmail.com); [malong@sdu.edu.cn](malong@sdu.edu.cn); [pfwang@sdu.edu.cn](pfwang@sdu.edu.cn); yfzhou @[sdu.edu.cn](sdu.edu.cn)). Ying He is with the School of Computer Science and Engineering, Nanyang Technological University, Singapore 639815 ([e-mail: yhe@ntu.edu.sg](e-mail: yhe@ntu.edu.sg)). Code is available at [https://github.com/yuanfeng-01/D-FRAME-test](https://github.com/yuanfeng-01/D-FRAME-test). Digital Object Identiﬁer 10.1109/TVCG.2025.3609350 (2) the complexity of geometric shapes, which often feature a high density of sharp edges in close proximity. Existing methods for extracting wireframes can be roughly divided into two categories: traditional geometric methods [1],[2],[3] and neural network-based methods [4],[5],[6],[7],[8],[9],[10] . Traditional geometric methods rely on the estimation ofcurvature or surface normal discontinuities to detect sharp features. Although effective for clean and dense data, these methods struggle with noise, sparsity, and complex geo

FAQs

What are the main challenges in extracting wireframes from CAD point clouds?

The paper identifies point-cloud defects such as lack of orientation, noise, and sparsity, and the geometric complexity of CAD shapes with densely packed sharp edges.

How does D-FRAME use a direction field to improve wireframe extraction?

D-FRAME introduces a direction field as auxiliary information to enhance edge detection accuracy, improve connectivity, and reduce errors caused by misclassified points.

How does D-FRAME address sparse or noisy edge points?

It applies a refinement strategy that smooths and completes the local edge point distribution before final wireframe construction.

Oct 26, 2026

Upload by WPS_1786070896