arxiv: 2602.21105 · v3 · submitted 2026-02-24 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

BrepGaussian: CAD reconstruction from Multi-View Images with Gaussian Splatting

Jiaxing Yu , Dongyang Ren , Hangyu Xu , Zhouyuxiao Yang , Yuanqi Li , Jie Guo , Zhengkang Zhou , Yanwen Guo

Authors on Pith no claims yet

Pith reviewed 2026-05-15 19:43 UTC · model grok-4.3

classification 💻 cs.CV

keywords B-Rep reconstructionGaussian SplattingCAD modelsMulti-view imagesParametric 3D representationComputer visionDeep learning for graphics

0 comments

The pith

BrepGaussian learns clean boundary representation models directly from multi-view 2D images using a two-stage Gaussian splatting process.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents BrepGaussian as a framework to recover explicit 3D boundary representations of solids from unstructured 2D images. It uses Gaussian splatting with learnable features in two stages: first to capture overall geometry and edges, then to refine features for coherent patches. This avoids reliance on dense point clouds that limit previous deep learning approaches and improves generalization to new shapes. A sympathetic reader would care because successful B-Rep recovery from images could streamline reverse engineering and CAD workflows in design and manufacturing.

Core claim

BrepGaussian is a novel framework that employs a Gaussian Splatting renderer with learnable features, followed by a specific fitting strategy in a two-stage learning process that first captures geometry and edges and then refines patch features to achieve clean geometry and coherent instance representations from multi-view images.

What carries the argument

The two-stage Gaussian Splatting framework with learnable features that disentangles geometry reconstruction from feature learning.

If this is right

Reconstructs B-Rep models without requiring dense and clean point cloud inputs.
Achieves cleaner geometry and more coherent instance representations compared to prior methods.
Generalizes better to novel shapes in CAD reconstruction tasks.
Demonstrates superior performance on standard benchmarks for multi-view 3D reconstruction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This approach might enable B-Rep extraction from real-world multi-view photos taken with consumer cameras.
The two-stage disentanglement could be adapted to other 3D representation learning problems beyond B-Rep.
Integration with existing CAD software pipelines could become feasible if the fitting strategy proves robust.

Load-bearing premise

The two-stage process reliably disentangles geometry reconstruction and feature learning from multi-view images to produce clean B-Rep geometry without post-hoc adjustments.

What would settle it

Experiments on a dataset of multi-view images showing frequent failures in producing valid manifold B-Rep models or requiring manual cleanup would falsify the claim of reliable reconstruction.

Figures

Figures reproduced from arXiv: 2602.21105 by Dongyang Ren, Hangyu Xu, Jiaxing Yu, Jie Guo, Yanwen Guo, Yuanqi Li, Zhengkang Zhou, Zhouyuxiao Yang.

**Figure 1.** Figure 1: Given multi-view images, our pipeline reconstructs a CAD model through feature-aware Gaussian Splatting and parametric [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Figure 2: Overall pipeline of BrepGaussian. Given multi-view RGB images of a CAD object, we extract edge and patch views using existing edge detection and segmentation models. These views drive a two-stage Gaussian Splatting model that predicts edge and patch labels on the reconstructed point cloud. The fitted primitives are globally optimized to obtain the final B-Rep model. Stage 2 training — learning 3D patch ins… view at source ↗

**Figure 3.** Figure 3: Illustration of optimized 2D Gaussians. flat regions use [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Qualitative comparison on patch segmentation. Our BrepGaussian produces cleaner and more consistent patch segmentation. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Qualitative comparison on CAD reconstruction. Our BrepGaussian exhibit more accurate geometry. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: From left to right, we present 2D images, point cloud [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 7.** Figure 7: Experiments on real-world scenes from ABO dataset [ [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

read the original abstract

The boundary representation (B-Rep) models a 3D solid as its explicit boundaries: trimmed corners, edges, and faces. Recovering B-Rep representation from unstructured data is a challenging and valuable task of computer vision and graphics. Recent advances in deep learning have greatly improved the recovery of 3D shape geometry, but still depend on dense and clean point clouds and struggle to generalize to novel shapes. We propose B-Rep Gaussian Splatting (BrepGaussian), a novel framework that learns 3D parametric representations from 2D images. We employ a Gaussian Splatting renderer with learnable features, followed by a specific fitting strategy. To disentangle geometry reconstruction and feature learning, we introduce a two-stage learning framework that first captures geometry and edges and then refines patch features to achieve clean geometry and coherent instance representations. Extensive experiments demonstrate the superior performance of our approach to state-of-the-art methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

BrepGaussian offers a two-stage Gaussian Splatting pipeline for direct B-Rep output from multi-view images, but the abstract supplies no metrics or details to support the superiority claim.

read the letter

The main takeaway is that this work tries to reconstruct clean B-Rep CAD models straight from images using Gaussian Splatting in two stages: first geometry and edges, then patch features. That combination is new relative to the point-cloud-heavy methods it cites, and it targets a practical need in graphics and industrial design where you want parametric output without dense intermediate clouds. The framing of disentangling geometry from features is reasonable on the surface and could help avoid messy post-processing. The paper does a decent job laying out why existing approaches struggle with generalization to novel shapes. The soft spots are the missing evidence. The abstract asserts better performance than state-of-the-art but gives no numbers, no datasets, no ablations, and no error analysis, so the central claim stays unverified. The stress-test point on feature leakage also lands: Gaussian primitives are continuous, and without shown loss terms, isolation mechanisms, or regularization to keep the first stage strictly geometric, the second stage risks producing non-clean edges and faces. That would undermine the no-post-hoc-adjustments premise. This is for readers working on image-to-CAD pipelines or hybrid vision-graphics methods who want to see the full experiments. It deserves peer review because the pipeline idea is distinct enough to merit referee scrutiny, even if heavy revision on the evaluation side is likely needed.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces BrepGaussian, a novel framework for reconstructing boundary representation (B-Rep) CAD models from multi-view 2D images. It employs Gaussian Splatting with learnable features in a two-stage learning process: the first stage captures geometry and edges, while the second refines patch features to produce clean geometry and coherent instance representations. The authors claim this yields superior performance over state-of-the-art methods without requiring dense point-cloud input.

Significance. If the two-stage disentanglement holds and produces clean B-Rep outputs, the work could advance CAD reconstruction in computer vision by enabling direct parametric recovery from images and improving generalization to novel shapes, addressing limitations of prior methods reliant on point clouds.

major comments (2)

[Abstract] Abstract: the central claim of 'superior performance' to state-of-the-art methods is unsupported by any quantitative metrics, dataset details, ablation studies, or error analysis, which is load-bearing for verifying the framework's effectiveness.
[Two-stage learning framework] Two-stage learning framework: no loss formulation, architectural isolation, or regularization is provided to enforce disentanglement between geometry/edge capture and patch feature refinement; Gaussian primitives are continuous and may entangle edge sharpness with surface features, risking non-clean B-Rep outputs without post-hoc adjustments.

minor comments (1)

[Method] The description of the 'specific fitting strategy' following Gaussian Splatting is vague and lacks equations or pseudocode for how trimmed faces/edges are obtained from the splatted representation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive feedback. We address each major comment below and will revise the manuscript to strengthen the presentation of results and technical details.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim of 'superior performance' to state-of-the-art methods is unsupported by any quantitative metrics, dataset details, ablation studies, or error analysis, which is load-bearing for verifying the framework's effectiveness.

Authors: We agree that the abstract would benefit from greater specificity. The full manuscript contains quantitative comparisons (including metrics such as Chamfer distance and edge accuracy), dataset descriptions, ablation studies, and error analysis in the Experiments section. In the revision we will expand the abstract to briefly cite the key performance gains and reference the supporting experiments, while preserving its concise nature. revision: yes
Referee: [Two-stage learning framework] Two-stage learning framework: no loss formulation, architectural isolation, or regularization is provided to enforce disentanglement between geometry/edge capture and patch feature refinement; Gaussian primitives are continuous and may entangle edge sharpness with surface features, risking non-clean B-Rep outputs without post-hoc adjustments.

Authors: We appreciate this observation. The manuscript describes the two-stage process at a high level; we will add the explicit loss functions for each stage (geometry/edge term in stage 1, patch-feature refinement term in stage 2), the architectural isolation mechanism (parameter freezing and separate optimizers), and any regularization terms used to promote disentanglement. We will also clarify how the subsequent B-Rep fitting strategy, combined with the learned edge features, produces clean parametric outputs and will include additional analysis or ablations demonstrating that entanglement is limited without relying on extensive post-processing. revision: yes

Circularity Check

0 steps flagged

No circularity: two-stage pipeline presented as independent without self-referential reductions

full rationale

The provided manuscript text describes a two-stage framework (first capturing geometry and edges via Gaussian Splatting, then refining patch features) and a fitting strategy to produce B-Rep outputs from multi-view images. No equations, loss formulations, or derivation steps are shown that reduce any claimed prediction or result to a fitted parameter or input quantity by construction. No self-citations are invoked as load-bearing uniqueness theorems, no ansatzes are smuggled via prior work, and no known results are merely renamed. The separation of stages is asserted as a design choice without mathematical reduction to the method's own outputs, leaving the derivation chain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the premise that Gaussian Splatting with learnable features plus a staged fitting procedure can recover explicit B-Rep boundaries from images alone; no free parameters or invented entities are named in the abstract.

axioms (1)

domain assumption Multi-view images contain enough geometric information to recover explicit trimmed faces, edges, and corners of a solid
Implicit in the task formulation and the choice of input modality.

pith-pipeline@v0.9.0 · 5480 in / 1140 out tokens · 18523 ms · 2026-05-15T19:43:21.699000+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We employ a Gaussian Splatting renderer with learnable features, followed by a specific fitting strategy. To disentangle geometry reconstruction and feature learning, we introduce a two-stage learning framework...
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Stage 1 Loss... Lstage1 = Lgeo + 0.1Ledge; Stage 2 Loss... Ltri with triplet margin

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

[1]

Gencad: Image-conditioned computer-aided design generation with transformer-based contrastive representation and diffusion priors.Trans

Md Ferdous Alam et al. Gencad: Image-conditioned computer-aided design generation with transformer-based contrastive representation and diffusion priors.Trans. Mach. Learn. Res., 2025. 3

work page 2025
[2]

Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P

Jonathan T. Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman, Ricardo Martin-Brualla, and Pratul P. Srinivasan. Mip-nerf: A multiscale representation for anti-aliasing neu- ral radiance fields. InInt. Conf. Comput. Vis., 2021. 3

work page 2021
[3]

Eulàlia Parés

Dena Bazazian and M. Eulàlia Parés. Edc-net: Edge detec- tion capsule network for 3d point clouds. InApplied Sci- ences, 2021. 2

work page 2021
[4]

Segment any 3d gaussians

Jiazhong Cen, Jiemin Fang, Chen Yang, Lingxi Xie, Xi- aopeng Zhang, Wei Shen, and Qi Tian. Segment any 3d gaussians. InAAAI, 2025. 3

work page 2025
[5]

Chen and et al

Y . Chen and et al. Capri-net: Learning compact cad shapes with adaptive primitive assembly. InIEEE Conf. Comput. Vis. Pattern Recog., 2022. 2

work page 2022
[6]

Ya- go Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik

Jasmine Collins, Shubham Goel, Kenan Deng, Achleshwar Luthra, Leon Xu, Erhan Gundogdu, Xi Zhang, Tomas F. Ya- go Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik. Abo: Dataset and bench- marks for real-world 3d object understanding. InIEEE Conf. Comput. Vis. Pattern Recog., 2022. 8

work page 2022
[7]

Fischler and Robert C

Martin A. Fischler and Robert C. Bolles. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography.Communications of the ACM, 1981. 5

work page 1981
[8]

Curve-aware gaussian splatting for 3d parametric curve reconstruction

Zhirui Gao, Renjiao Yi, Yaqiao Dai, Xuening Zhu, Wei Chen, Chenyang Zhu, and Kai Xu. Curve-aware gaussian splatting for 3d parametric curve reconstruction. InInt. Conf. Comput. Vis., 2025. 3

work page 2025
[9]

Self-supervised learning of hy- brid part-aware 3d representations of 2d gaussians and su- perquadrics

Zhirui Gao, Renjiao Yi, Yuhang Huang, Wei Chen, Chenyang Zhu, and Kai Xu. Self-supervised learning of hy- brid part-aware 3d representations of 2d gaussians and su- perquadrics. InInt. Conf. Comput. Vis., 2025. 3

work page 2025
[10]

Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering

Antoine Guédon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering. InIEEE Conf. Com- put. Vis. Pattern Recog., 2024. 3

work page 2024
[11]

Complexgen: Cad reconstruction by b- rep chain complex generation

Haoxiang Guo, Shilin Liu, Hao Pan, Yang Liu, Xin Tong, and Baining Guo. Complexgen: Cad reconstruction by b- rep chain complex generation. InACM SIGGRAPH Annual Conference, 2022. 2

work page 2022
[12]

Semantic gaussians: Open-vocabulary scene understanding with 3d gaussian splatting.ArXiv, abs/2403.15624, 2024

Jun Guo, Xiaojian Ma, Yue Fan, Huaping Liu, and Qing Li. Semantic gaussians: Open-vocabulary scene understanding with 3d gaussian splatting.ArXiv, abs/2403.15624, 2024. 3

work page arXiv 2024
[13]

2d gaussian splatting for geometrically ac- curate radiance fields

Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. InACM SIGGRAPH Annual Confer- ence, 2024. 2

work page 2024
[14]

Ucsg-net: Unsupervised discovering of constructive solid geometry tree

Kacper Kania, Maciej Zi˛ eba, and Tomasz Kajdanowicz. Ucsg-net: Unsupervised discovering of constructive solid geometry tree. InAdv. Neural Inform. Process. Syst., 2020. 2

work page 2020
[15]

3d gaussian splatting for real-time radiance field rendering

Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. InACM SIGGRAPH Annual Con- ference, 2023. 2, 3

work page 2023
[16]

Berg, Wan-Yen Lo, Piotr Dollar, and Ross Girshick

Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C. Berg, Wan-Yen Lo, Piotr Dollar, and Ross Girshick. Segment anything. InInt. Conf. Comput. Vis., 2023. 3, 5

work page 2023
[17]

Abc: A big cad model dataset for geometric deep learning

Sebastian Koch, Albert Matveev, Zhongshi Jiang, Francis Williams, Alexey Artemov, Evgeny Burnaev, Marc Alexa, Denis Zorin, and Daniele Panozzo. Abc: A big cad model dataset for geometric deep learning. InIEEE Conf. Comput. Vis. Pattern Recog., 2019. 5

work page 2019
[18]

Lambourne, Karl D

Joseph G. Lambourne, Karl D. D. Willis, Pradeep Ku- mar Jayaraman, Aditya Sanghi, Peter Meltzer, and Hooman Shayani. Brepnet: A topological message passing system for solid models. InIEEE Conf. Comput. Vis. Pattern Recog.,

work page
[19]

Eric-Tuan Lê, Minhyuk Sung, Duygu Ceylan, Radomír Mˇech, Tamy Boubekeur, and Niloy J. Mitra. Cpfn: Cascaded primitive fitting networks for high-resolution point clouds. In Int. Conf. Comput. Vis., 2021. 2

work page 2021
[20]

Supervised fitting of geometric primi- tives to 3d point clouds

Lingxiao Li, Minhyuk Sung, Anastasia Dubrovina, Li Yi, and Leonidas Guibas. Supervised fitting of geometric primi- tives to 3d point clouds. InIEEE Conf. Comput. Vis. Pattern Recog., 2019. 1, 2

work page 2019
[21]

Secad-net: Self-supervised cad reconstruction by learning sketch-extrude operations

Pu Li, Jianwei Guo, Xiaopeng Zhang, and Dong-Ming Yan. Secad-net: Self-supervised cad reconstruction by learning sketch-extrude operations. InIEEE Conf. Comput. Vis. Pat- tern Recog., 2023. 2

work page 2023
[22]

Sfmcad: Unsupervised cad reconstruction by learning sketch-based feature modeling operations

Pu Li, Jianwei Guo, Huibin Li, Bedrich Benes, and Dong- Ming Yan. Sfmcad: Unsupervised cad reconstruction by learning sketch-based feature modeling operations. InIEEE Conf. Comput. Vis. Pattern Recog., 2024. 2

work page 2024
[23]

Yangyan Li, Xiaokun Wu, Yiorgos Chrysathou, Andrei Sharf, Daniel Cohen-Or, and Niloy J. Mitra. Globfit: Consis- tently fitting primitives by discovering global relations.ACM Trans. Graph., 2011. 2

work page 2011
[24]

Sed-net: Surface and edge detection for primitive fitting of point clouds

Yuanqi Li, Shun Liu, Xinran Yang, Jianwei Guo, Jie Guo, and Yanwen Guo. Sed-net: Surface and edge detection for primitive fitting of point clouds. InACM SIGGRAPH Annual Conference, 2023. 2, 7

work page 2023
[25]

Caddreamer: Cad object generation from single-view images

Yuan Li, Cheng Lin, Yuan Liu, Xiaoxiao Long, Chenxu Zhang, Ningna Wang, Xin Li, Wenping Wang, and Xiaohu Guo. Caddreamer: Cad object generation from single-view images. InIEEE Conf. Comput. Vis. Pattern Recog., 2025. 3

work page 2025
[26]

Deep point cloud edge recon- struction via surface patch segmentation.IEEE Transactions on Visualization and Computer Graphics, 2025

Yuanqi Li, Hongshen Wang, Yansong Liu, Jingcheng Huang, Shun Liu, and Chenyu Huang. Deep point cloud edge recon- struction via surface patch segmentation.IEEE Transactions on Visualization and Computer Graphics, 2025. 2, 7

work page 2025
[27]

Split-and-fit: Learning b-reps via structure-aware voronoi partitioning

Yilin Liu, Jiale Chen, Shanshan Pan, Daniel Cohen-Or, Hao Zhang, and Hui Huang. Split-and-fit: Learning b-reps via structure-aware voronoi partitioning. InACM SIGGRAPH Annual Conference, 2024. 2, 7, 8

work page 2024
[28]

Point2cad: Reverse engineering cad models from 3d point clouds

Yujia Liu, Anton Obukhov, Jan Dirk Wegner, and Konrad Schindler. Point2cad: Reverse engineering cad models from 3d point clouds. InIEEE Conf. Comput. Vis. Pattern Recog.,

work page
[29]

Hola: B-rep generation using a holistic latent representation

Yilin Liu et al. Hola: B-rep generation using a holistic latent representation. InSIGGRAPH, 2025. 3

work page 2025
[30]

Def: Deep es- timation of sharp geometric features in 3d shapes

Albert Matveev, Ruslan Rakhimov, Alexey Artemov, Gleb Bobrovskikh, Vage Egiazarian, Emil Bogomolov, Daniele Panozzo, Denis Zorin, and Evgeny Burnaev. Def: Deep es- timation of sharp geometric features in 3d shapes. InACM SIGGRAPH Annual Conference, 2022. 2

work page 2022
[31]

Srinivasan, Matthew Tancik, Jonathan T

Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. InEur. Conf. Comput. Vis., 2020. 2, 3

work page 2020
[32]

Instant neural graphics primitives with a multires- olution hash encoding

Thomas Müller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a multires- olution hash encoding. InACM SIGGRAPH Annual Confer- ence, pages 102:1–102:15, 2022. 3

work page 2022
[33]

Soft 3d reconstruction for view synthesis

Eric Penner and Li Zhang. Soft 3d reconstruction for view synthesis. InACM Trans. Graph., 2017. 3

work page 2017
[34]

Csg-stump: A learning friendly csg-like representation for interpretable shape parsing

Daxuan Ren, Jianmin Zheng, Jianfei Cai, Jiatong Li, Haiyong Jiang, Zhongang Cai, Junzhe Zhang, Liang Pan, Mingyuan Zhang, Haiyu Zhao, and Shuai Yi. Csg-stump: A learning friendly csg-like representation for interpretable shape parsing. InInt. Conf. Comput. Vis., 2021. 2

work page 2021
[35]

Csgnet: Neural shape parser for constructive solid geometry

Gopal Sharma, Rishabh Goyal, Difan Liu, Evangelos Kalogerakis, and Subhransu Maji. Csgnet: Neural shape parser for constructive solid geometry. InIEEE Conf. Com- put. Vis. Pattern Recog., 2018. 2

work page 2018
[36]

Parsenet: A parametric surface fitting network for 3d point clouds

Gopal Sharma, Difan Liu, Subhransu Maji, Evangelos Kalogerakis, Siddhartha Chaudhuri, and Radomír M ˇech. Parsenet: A parametric surface fitting network for 3d point clouds. InEur. Conf. Comput. Vis., 2020. 1, 2, 7

work page 2020
[37]

Point2cyl: Reverse engineering 3d objects from point clouds to extrusion cylinders

Mikaela Angelina Uy, Yen-Yu Chang, Minhyuk Sung, Purvi Goel, Joseph Lambourne, Tolga Birdal, and Leonidas Guibas. Point2cyl: Reverse engineering 3d objects from point clouds to extrusion cylinders. InIEEE Conf. Comput. Vis. Pattern Recog., 2022. 2

work page 2022
[38]

Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. InAdv. Neural Inform. Process. Syst., 2021. 3

work page 2021
[39]

Pie-net: Parametric inference of point cloud edges

Xiaogang Wang, Yuelang Xu, Kai Xu, Andrea Tagliasacchi, Bin Zhou, Ali Mahdavi-Amiri, and Hao Zhang. Pie-net: Parametric inference of point cloud edges. InAdv. Neural Inform. Process. Syst., 2020. 2

work page 2020
[40]

Hpnet: Deep primi- tive segmentation using hybrid representations

Siming Yan, Zhenpei Yang, Chongyang Ma, Haibin Huang, Etienne V ouga, and Qixing Huang. Hpnet: Deep primi- tive segmentation using hybrid representations. InInt. Conf. Comput. Vis., 2021. 2, 7

work page 2021
[41]

Sgcr: Spherical gaussians for efficient 3d curve reconstruction

Xinran Yang, Donghao Ji, Yuanqi Li, Jie Guo, Yanwen Guo, and Junyuan Xie. Sgcr: Spherical gaussians for efficient 3d curve reconstruction. InIEEE Conf. Comput. Vis. Pattern Recog., 2025. 3

work page 2025
[42]

Nef: Neural edge fields for 3d parametric curve reconstruction from multi-view images

Yunfan Ye, Renjiao Yi, Zhirui Gao, Chenyang Zhu, Zhiping Cai, and Kai Xu. Nef: Neural edge fields for 3d parametric curve reconstruction from multi-view images. InIEEE Conf. Comput. Vis. Pattern Recog., 2023. 5

work page 2023
[43]

Sketchsplat: 3d edge reconstruction via differentiable multi-view sketch splatting

Haiyang Ying and Matthias Zwicker. Sketchsplat: 3d edge reconstruction via differentiable multi-view sketch splatting. InInt. Conf. Comput. Vis., 2025. 3

work page 2025
[44]

Img2cad: Re- verse engineering 3d cad models from images through vlm- assisted conditional factorization

Yang You, Mikaela Angelina Uy, Jiaqi Han, Rahul Thomas, Haotong Zhang, Yi Du, Hansheng Chen, Francis Engel- mann, Suya You, and Leonidas Guibas. Img2cad: Re- verse engineering 3d cad models from images through vlm- assisted conditional factorization. InSIGGRAPH Asia, 2025. 3

work page 2025
[45]

Dualcsg: Learning dual csg trees for general and compact cad modeling.ArXiv, abs/2301.11497,

Fenggen Yu, Qimin Chen, Maham Tanveer, Ali Mahdavi- Amiri, and Hao Zhang. Dualcsg: Learning dual csg trees for general and compact cad modeling.ArXiv, abs/2301.11497,

work page arXiv
[46]

Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields

Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Ze- hao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, and Achuta Kadambi. Feature 3dgs: Supercharging 3d gaussian splatting to enable distilled feature fields. In IEEE Conf. Comput. Vis. Pattern Recog., 2023. 3

work page 2023
[47]

Nerve: Neural volumetric edges for parametric curve extraction from point clouds

Xiangyu Zhu, Dong Du, Weikai Chen, Zhiyou Zhao, Yinyu Nie, and Xiaoguang Han. Nerve: Neural volumetric edges for parametric curve extraction from point clouds. InIEEE Conf. Comput. Vis. Pattern Recog., 2023. 2

work page 2023