arxiv: 2605.14029 · v1 · submitted 2026-05-13 · 💻 cs.GR · cs.CG

Recognition: 2 theorem links

· Lean Theorem

Fast and Robust Mesh Simplification for Generated and Real-World 3D Assets

Kunal Bhosikar , Preet Savalia , Lokender Tiwari , Brojeshwar Bhowmick

Authors on Pith no claims yet

Pith reviewed 2026-05-15 05:45 UTC · model grok-4.3

classification 💻 cs.GR cs.CG

keywords mesh simplificationquadric error metricfeature preservation3D reconstructionAI-generated meshestexture mappingnon-manifold meshes

0 comments

The pith

A multi-term quadric error metric simplifies noisy 3D meshes faster while preserving sharp features better than standard approaches.

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

The paper presents Feature-Aware Quadric Error Metric (FA-QEM) as a mesh simplification pipeline for dense, noisy meshes produced by neural rendering and large-scale 3D generation. It replaces the classic quadric error with a combined formulation that adds boundary curvature and surface normal consistency to the usual geometric deviation term. This joint encoding guides vertex placement so that sharp edges survive even when triangle counts drop by large factors. The method runs faster than prior techniques and produces simplified meshes that support higher-quality texture transfer in later steps. Results on AI-generated models and real datasets such as Thingi10K confirm lower geometric error and greater robustness to non-manifold inputs.

Core claim

The paper claims that a novel multi-term quadric error formulation jointly encoding geometric deviation, boundary curvature, and surface normal consistency enables optimal vertex placement that preserves sharp features even under aggressive simplification, yielding lower error, better visual fidelity, and faster runtimes than existing methods on both generated and real-world meshes.

What carries the argument

The multi-term quadric error formulation that jointly encodes geometric deviation, boundary curvature, and surface normal consistency to determine vertex placement.

If this is right

Simplified meshes retain sharp edges and fine structures at high reduction ratios.
Downstream texture mapping and appearance transfer achieve higher fidelity than with meshes simplified by prior methods.
Runtime remains lower while robustness holds across non-manifold and noisy inputs.
Geometric error metrics stay consistently below those of standard quadric approaches on the same target complexity.
The pipeline integrates directly as a front-end step in reconstruction and generation workflows.

Where Pith is reading between the lines

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

Early insertion of this simplification step could reduce memory and compute demands in real-time AR/VR rendering loops.
The same multi-term idea might extend to related tasks such as mesh compression or level-of-detail generation.
Improved front-end meshes may lower the cost of accurate physics simulation on generated 3D content.
Further tests on outputs from newer generative models could show whether the added terms remain effective under extreme noise.

Load-bearing premise

That adding boundary curvature and normal consistency terms to the quadric error will produce measurably better vertex placement and feature preservation than the standard single-term version without creating new artifacts or requiring per-mesh parameter tuning.

What would settle it

A direct comparison on a fixed set of noisy non-manifold meshes where FA-QEM at a target face count yields higher Hausdorff distance or visibly more feature loss than the baseline quadric error metric.

Figures

Figures reproduced from arXiv: 2605.14029 by Brojeshwar Bhowmick, Kunal Bhosikar, Lokender Tiwari, Preet Savalia.

**Figure 1.** Figure 1: FA-QEM delivers high-fidelity and efficient mesh simplification. We compare against Liu et al. (STMW) [17], QEM4VR [1], QEM [5], RobustLPM [2], CWF [30], and ICE [16] at 10% resolution. F and T denote face count and runtime (s). Close-ups show that FA-QEM preserves sharp geometry and fine textures under aggressive simplification while achieving significantly lower runtimes. Several baselines do not support… view at source ↗

**Figure 2.** Figure 2: Overview of the FA-QEM pipeline. We construct a composite quadric Qgf from base, boundary–curvature, and normalalignment terms to define costgf , alongside an area-preservation quadric Q A with cost costarea. Edge collapses are guided by a weighted combination yielding costtotal. After simplification, successive mapping ensures consistent, high-fidelity texture transfer. See Sec. 3.2 for details. where Kp… view at source ↗

**Figure 3.** Figure 3: Key mesh terminology. A boundary edge is incident to one face, a non-manifold edge is shared by more than two faces, and a virtual edge is an artificial collapse candidate used to merge nearby disconnected components [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Qualitative comparison with state-of-the-art methods. We compare geometric and textured meshes simplified to a fixed face count F. T denotes runtime (s), and a red X indicates failure to reach the target resolution. Top two rows are from the Real-World Textured Things dataset [21], and the bottom row shows an AI-generated mesh [33]. FA-QEM preserves fine geometry and texture under aggressive simplification… view at source ↗

**Figure 5.** Figure 5: Sensitivity Analysis: wboundary. The error drops to a clear minimum around 500. The stable “U-shape” confirms the parameter is well-tuned and robust. 0 50 100 150 200 250 300 warea Parameter Value 14 15 16 17 18 19 H a u s d o r ff Dis t a n c e (× 1 0 2 ) 18 20 22 24 26 C h a m fe r Dis t a n c e (× 1 0 3 ) Sensitivity Analysis for warea Hyperparameter Hausdorff Error Our Chosen Value (100) Chamfer Error … view at source ↗

**Figure 6.** Figure 6: Sensitivity Analysis: warea. Introduction of the area term significantly reduces error compared to the baseline (0), with stable performance observed for values above 50. remains stable within a reasonable range. Inverse Area Weighting (wplane area) [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

**Figure 7.** Figure 7: Sensitivity Analysis: wnormal. A small weight of 0.01 provides the optimal balance, reducing faceting without overconstraining the geometry. 0.0 0.5 1.0 1.5 2.0 2.5 3.0 wplanearea Parameter Value 12.25 12.50 12.75 13.00 13.25 13.50 13.75 14.00 H a u s d o r ff Dis t a n c e (× 1 0 2 ) 13.5 14.0 14.5 15.0 15.5 16.0 16.5 17.0 C h a m fe r Dis t a n c e (× 1 0 3 ) Sensitivity Analysis for wplanearea Hyperpar… view at source ↗

**Figure 8.** Figure 8: Sensitivity Analysis: wplane area. The inverse-area weighting effectively guides simplification to flat regions, minimizing error at our chosen value of 1.0. decoupled, 2-stage cost metric. 1. Efficient Optimal Position Search: By creating a single, composite geometric-feature quadric (Qgf ) upfront, we only need to solve one small (3×3) linear system per edge to find the optimal vertex position. This is… view at source ↗

**Figure 9.** Figure 9: Runtime Scalability Analysis. We plot the execution time of FA-QEM against input mesh size on a standard Intel Core i5 CPU. The method exhibits consistent polynomial scaling (O(N 2 )), verifying its stability across varying mesh complexities. Notably, for the high-resolution ‘Lamp’ model (240k faces), FAQEM completes in 28 minutes, achieving a 14.7× speedup over the state-of-the-art robust method Liu et … view at source ↗

**Figure 11.** Figure 11: We present additional qualitative simplification results at 50% resolution of examples shown in our main paper Fig [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗

**Figure 12.** Figure 12: Sample qualitative results at 50% resolution: We present additional results 50% resolution. The first example is from RealWorld Textured Things dataset [21], while the last two examples are AI generated using Hunyuan3D [33]. The results on 10% resolution of the these examples are shown in [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗

**Figure 13.** Figure 13: Sample qualitative results at 10% resolution: Results on 10% resolution of the examples shown in [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗

read the original abstract

The rapid growth of 3D content from modern reconstruction and generative pipelines, such as neural rendering and large-scale 3D asset generation, has led to an abundance of dense, noisy, and often non-manifold meshes. While these representations achieve high visual fidelity, their complexity poses significant challenges for downstream applications in simulation, AR/VR, and scientific computing, where efficient and reliable geometry is essential. This necessitates mesh simplification methods that are not only fast and robust to "in-the-wild" inputs, but also capable of preserving fine geometric structures and high-quality appearance. In this paper, we propose Feature-Aware Quadric Error Metric (FA-QEM), a comprehensive mesh simplification pipeline designed for modern 3D assets. Our approach introduces a novel multi-term quadric error formulation that jointly encodes geometric deviation, boundary curvature, and surface normal consistency, enabling optimal vertex placement that preserves sharp features even under aggressive simplification. Furthermore, we show that high-fidelity geometric simplification significantly improves downstream appearance transfer, serving as a superior front-end for texture mapping via successive mapping techniques. We conduct extensive evaluations on both AI-generated meshes and large-scale real-world datasets, including Thingi10K and the Real-World Textured Things dataset. Our results demonstrate that FA-QEM achieves consistently lower geometric error, better visual fidelity, and substantially faster runtimes compared to existing methods, while maintaining robustness across diverse and challenging inputs. These properties make FA-QEM a practical and effective component for scalable 3D reconstruction and generation pipelines.

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

FA-QEM folds boundary curvature and normal consistency into the classic quadric error metric to handle noisy generated meshes, with reported speed and fidelity gains that look useful but incremental.

read the letter

The core update is a multi-term quadric that adds boundary curvature and normal consistency to the usual geometric deviation term. This is meant to keep sharp features intact during aggressive simplification of dense, noisy meshes from generative pipelines. The abstract positions it as a drop-in improvement that also helps downstream texture mapping. Evaluations on Thingi10K and the Real-World Textured Things dataset are the strongest part: they report lower geometric error, better visual quality, and faster runtimes than existing methods, with claims of robustness across AI-generated and real inputs. That combination of speed and feature preservation is the practical selling point for anyone running simplification in AR/VR or simulation loops. The formulation itself is a straightforward extension of QEM rather than a new theoretical framework, so the novelty sits in the specific term combination and the end-to-end pipeline. On the soft spots, the stress-test concern about weights is worth checking in the full text. If the three terms use fixed global coefficients that hold up without per-mesh retuning, the robustness claim is solid; if the reported gains shrink or artifacts appear under strict fixed weights, the advantage narrows. The abstract does not spell out the exact weighting scheme or optimization details, so the experiments need to show that the vertex placement step actually delivers the claimed feature retention without introducing new problems. This paper is for graphics practitioners who need a reliable, fast simplification step for in-the-wild 3D assets. A reader already working with QEM-based tools would get immediate value from the implementation choices and the downstream appearance-transfer results. It is not reshaping the field, but it addresses a real engineering pain point with concrete numbers. I would send it to peer review because the experiments target relevant datasets and the method is grounded enough to benefit from referee feedback on the weighting and comparison details.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Feature-Aware Quadric Error Metric (FA-QEM), a mesh simplification pipeline that augments standard quadric error metrics with additional boundary-curvature and surface-normal-consistency terms. The central claim is that this multi-term formulation enables optimal vertex placement that preserves sharp features under aggressive simplification of dense, noisy, non-manifold meshes from generative and real-world sources, while delivering lower geometric error, higher visual fidelity, faster runtimes, and improved downstream texture mapping. Experiments are reported on Thingi10K and the Real-World Textured Things dataset, with comparisons to existing simplification methods.

Significance. If the added terms with fixed, mesh-independent weights produce consistent, artifact-free gains over plain QEM across noisy generated meshes, the work would offer a practical, tuning-free front-end for 3D asset pipelines. The reported speed advantage and downstream appearance-transfer benefit would be valuable for scalable reconstruction and generation workflows. The extensive dataset coverage is a strength.

major comments (2)

[Section 3] Section 3 (multi-term quadric error formulation): the manuscript must explicitly state the fixed global weights applied to the geometric-deviation, boundary-curvature, and normal-consistency terms and provide an ablation confirming that these weights remain effective without per-mesh adjustment on the diverse noisy inputs; otherwise the robustness and “no tuning” claims rest on an unverified assumption.
[Table 2] Table 2 (quantitative results on Thingi10K): the reported geometric-error reductions versus standard QEM and other baselines should include per-mesh standard deviations or statistical tests; without them it is unclear whether the observed improvements are consistent or driven by a few favorable cases.

minor comments (2)

[Figure 4] Figure 4 (visual comparisons): the simplification ratios and camera angles should be stated explicitly in the caption so readers can directly compare feature preservation.
[Section 4.3] Section 4.3 (downstream texture mapping): clarify whether the reported appearance-transfer gains are measured with the same mapping algorithm across all simplification methods or whether FA-QEM meshes receive additional post-processing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Section 3] Section 3 (multi-term quadric error formulation): the manuscript must explicitly state the fixed global weights applied to the geometric-deviation, boundary-curvature, and normal-consistency terms and provide an ablation confirming that these weights remain effective without per-mesh adjustment on the diverse noisy inputs; otherwise the robustness and “no tuning” claims rest on an unverified assumption.

Authors: We agree that explicit statement of the weights and supporting evidence are necessary. In the revised manuscript we will state the fixed global weights used for the geometric-deviation, boundary-curvature, and normal-consistency terms directly in Section 3. We will also add an ablation study that evaluates these same fixed weights on the full range of noisy inputs from Thingi10K and the Real-World Textured Things dataset, confirming that no per-mesh retuning is required. revision: yes
Referee: [Table 2] Table 2 (quantitative results on Thingi10K): the reported geometric-error reductions versus standard QEM and other baselines should include per-mesh standard deviations or statistical tests; without them it is unclear whether the observed improvements are consistent or driven by a few favorable cases.

Authors: We accept that reporting variability is important for demonstrating consistency. We will revise Table 2 to include per-mesh standard deviations of the geometric-error metrics, thereby allowing readers to evaluate whether the reported gains hold across the dataset rather than being driven by outliers. revision: yes

Circularity Check

0 steps flagged

No circularity: FA-QEM extends external QEM with independent additive terms

full rationale

The central derivation adds boundary-curvature and normal-consistency terms to the established Garland-Heckbert quadric error metric. These terms are defined directly from mesh geometry (curvature along boundaries, normal deviation) rather than fitted to the simplification output or derived from self-citations. No equation reduces the claimed improvement to a parameter defined by the result itself, and the paper cites the original QEM work as an external foundation. Evaluations on Thingi10K and Real-World Textured Things datasets supply independent empirical checks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on extending the standard quadric error metric with two new penalty terms whose effectiveness is asserted but not derived from first principles.

axioms (1)

domain assumption A linear combination of geometric, boundary, and normal terms yields an error metric whose minimization produces optimal vertex placement for feature preservation
Invoked when the paper states the multi-term formulation enables optimal placement

pith-pipeline@v0.9.0 · 5589 in / 1153 out tokens · 31425 ms · 2026-05-15T05:45:30.406209+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/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

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

We propose a novel multi-term quadric error formulation that jointly encodes geometric deviation, boundary curvature, and surface normal consistency... Qk_gf = Qk_base + Qk_boundary + Qk_normal (Eq. 4) with fixed weights w_area=100, w_boundary=500, w_normal=0.01, w_plane_area=1.0 (Table 1).
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

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

cost_total(v′) = cost_gf(v′) + w_area · cost_area(v′) (Eq. 2); boundary curvature via discrete κ and dual-plane quadrics (Eqs. 6-9).

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

41 extracted references · 41 canonical work pages

[1]

Designing and evaluating a mesh simplification algorithm for virtual reality.ACM Transac- tions on Multimedia Computing, Communications, and Ap- plications, 14:1–26, 2018

Kanchan Bahirat, Chengyuan Lai, Ryan McMahan, and Bal- akrishnan Prabhakaran. Designing and evaluating a mesh simplification algorithm for virtual reality.ACM Transac- tions on Multimedia Computing, Communications, and Ap- plications, 14:1–26, 2018. 1, 2, 4, 5, 6, 7

work page 2018
[2]

Robust low-poly meshing for general 3d mod- els.ACM Trans

Zhen Chen, Zherong Pan, Kui Wu, Etienne V ouga, and Xifeng Gao. Robust low-poly meshing for general 3d mod- els.ACM Trans. Graph., 42(4), 2023. 1, 2, 6

work page 2023
[3]

Cignoni, C

P. Cignoni, C. Montani, and R. Scopigno. A comparison of mesh simplification algorithms.Computers & Graphics, 22 (1):37–54, 1998. 2

work page 1998
[4]

Garland and P.S

M. Garland and P.S. Heckbert. Simplifying surfaces with color and texture using quadric error metrics. InProceed- ings Visualization ’98 (Cat. No.98CB36276), pages 263– 269, 1998. 2, 4

work page 1998
[5]

Heckbert.Surface Simplifica- tion Using Quadric Error Metrics

Michael Garland and Paul S. Heckbert.Surface Simplifica- tion Using Quadric Error Metrics. Association for Comput- ing Machinery, New York, NY , USA, 1 edition, 2023. 1, 2, 3, 6, 7, 5

work page 2023
[6]

Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering, 2023

Antoine Gu ´edon and Vincent Lepetit. Sugar: Surface- aligned gaussian splatting for efficient 3d mesh reconstruc- tion and high-quality mesh rendering, 2023. 1

work page 2023
[7]

A data reduction scheme for triangulated surfaces.Computer Aided Geometric Design, 11(2):197– 214, 1994

Bernd Hamann. A data reduction scheme for triangulated surfaces.Computer Aided Geometric Design, 11(2):197– 214, 1994. 2

work page 1994
[8]

Meshcnn: a network with an edge.ACM Transactions on Graphics, 38(4):1–12, 2019

Rana Hanocka, Amir Hertz, Noa Fish, Raja Giryes, Shachar Fleishman, and Daniel Cohen-Or. Meshcnn: a network with an edge.ACM Transactions on Graphics, 38(4):1–12, 2019. 2

work page 2019
[9]

Progressive meshes

Hugues Hoppe. Progressive meshes. InProceedings of the 23rd Annual Conference on Computer Graphics and Inter- active Techniques, page 99–108, New York, NY , USA, 1996. Association for Computing Machinery. 2

work page 1996
[10]

H. Hoppe. New quadric metric for simplifying meshes with appearance attributes. InProceedings Visualization ’99 (Cat. No.99CB37067), pages 59–510, 1999. 2

work page 1999
[11]

Mesh optimization

Hugues Hoppe, Tony DeRose, Tom Duchamp, John McDon- ald, and Werner Stuetzle. Mesh optimization. InProceedings of the 20th Annual Conference on Computer Graphics and Interactive Techniques, page 19–26, New York, NY , USA,

work page
[13]

Robust inside-outside segmentation using generalized winding num- bers.ACM Trans

Alec Jacobson, Ladislav Kavan, and Olga Sorkine. Robust inside-outside segmentation using generalized winding num- bers.ACM Trans. Graph., 32(4), 2013. 5

work page 2013
[14]

3d gaussian splatting for real-time radiance field rendering, 2023

Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering, 2023. 1, 2

work page 2023
[15]

Lp centroidal voronoi tessellation and its applications.ACM Trans

Bruno L ´evy and Yang Liu. Lp centroidal voronoi tessellation and its applications.ACM Trans. Graph., 29(4), 2010. 2

work page 2010
[16]

Lindstrom and G

P. Lindstrom and G. Turk. Fast and memory efficient polyg- onal simplification. InProceedings Visualization ’98 (Cat. No.98CB36276), pages 279–286, 1998. 2, 4

work page 1998
[17]

Surface simplification using intrinsic error metrics, 2023

Hsueh-Ti Derek Liu, Mark Gillespie, Benjamin Chislett, Nicholas Sharp, Alec Jacobson, and Keenan Crane. Surface simplification using intrinsic error metrics, 2023. 1, 2, 6, 7

work page 2023
[18]

Sim- plifying triangle meshes in the wild, 2024

Hsueh-Ti Derek Liu, Xiaoting Zhang, and Cem Yuksel. Sim- plifying triangle meshes in the wild, 2024. 1, 2, 5, 6, 7, 8, 3, 4

work page 2024
[19]

Lorensen and Harvey E

William E. Lorensen and Harvey E. Cline. Marching cubes: A high resolution 3d surface construction algorithm.SIG- GRAPH Comput. Graph., 21(4):163–169, 1987. 1, 2

work page 1987
[20]

Model simplification us- ing vertex-clustering

Kok-Lim Low and Tiow-Seng Tan. Model simplification us- ing vertex-clustering. InProceedings of the 1997 Sympo- sium on Interactive 3D Graphics, page 75–ff., New York, NY , USA, 1997. Association for Computing Machinery. 2

work page 1997
[21]

D.P. Luebke. A developer’s survey of polygonal simplifica- tion algorithms.IEEE Computer Graphics and Applications, 21(3):24–35, 2001. 2

work page 2001
[22]

Real-world textured things: A repos- itory of textured models generated with modern photo- reconstruction tools.Computer Aided Geometric Design, 83: 101943, 2020

Andrea Maggiordomo, Federico Ponchio, Paolo Cignoni, and Marco Tarini. Real-world textured things: A repos- itory of textured models generated with modern photo- reconstruction tools.Computer Aided Geometric Design, 83: 101943, 2020. 6, 7, 1, 2, 3, 5

work page 2020
[23]

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, 2020. 1, 2

work page 2020
[24]

Barron, and Ben Milden- hall

Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion, 2022. 1

work page 2022
[25]

Multi-resolution 3d ap- proximation for rendering complex scenes

Jarek Rossignac and Paul Borrel. Multi-resolution 3d ap- proximation for rendering complex scenes. pages 455–465,

work page
[26]

Schroeder, Jonathan A

William J. Schroeder, Jonathan A. Zarge, and William E. Lorensen. Decimation of triangle meshes. InProceedings of the 19th Annual Conference on Computer Graphics and Interactive Techniques, page 65–70, New York, NY , USA,

work page
[28]

Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3d shape synthesis, 2021

Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra: a hybrid represen- tation for high-resolution 3d shape synthesis, 2021. 2

work page 2021
[29]

Trettner and L

P. Trettner and L. Kobbelt. Fast and robust qef minimization using probabilistic quadrics.Computer Graphics Forum, 39 (2):325–334, 2020. 2

work page 2020
[30]

Re-tiling polygonal surfaces

Greg Turk. Re-tiling polygonal surfaces. InProceedings of the 19th Annual Conference on Computer Graphics and Interactive Techniques, page 55–64, New York, NY , USA,

work page
[31]

Association for Computing Machinery. 2

work page
[32]

Structured 3d latents for scalable and versatile 3d gen- eration

Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d gen- eration. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 21469–21480, 2025. 6

work page 2025
[33]

Cwf: Consolidating weak features in high-quality mesh simplification, 2024

Rui Xu, Longdu Liu, Ningna Wang, Shuangmin Chen, Shiqing Xin, Xiaohu Guo, Zichun Zhong, Taku Komura, Wenping Wang, and Changhe Tu. Cwf: Consolidating weak features in high-quality mesh simplification, 2024. 1, 2, 6, 7

work page 2024
[34]

Pcn: Point completion network, 2019

Wentao Yuan, Tejas Khot, David Held, Christoph Mertz, and Martial Hebert. Pcn: Point completion network, 2019. 7

work page 2019
[35]

Mesh color textures

Cem Yuksel. Mesh color textures. InProceedings of High Performance Graphics, New York, NY , USA, 2017. Associ- ation for Computing Machinery. 5

work page 2017
[36]

Hunyuan3d 2.0: Scal- ing diffusion models for high resolution textured 3d assets generation, 2025

Zibo Zhao, Zeqiang Lai, Qingxiang Lin, Yunfei Zhao, Haolin Liu, Shuhui Yang, Yifei Feng, Mingxin Yang, Sheng Zhang, Xianghui Yang, Huiwen Shi, Sicong Liu, Junta Wu, Yihang Lian, Fan Yang, Ruining Tang, Zebin He, Xinzhou Wang, Jian Liu, Xuhui Zuo, Zhuo Chen, Biwen Lei, Hao- han Weng, Jing Xu, Yiling Zhu, Xinhai Liu, Lixin Xu, Changrong Hu, Shaoxiong Yang, ...

work page 2025
[37]

Thingi10k: A dataset of 10,000 3d-printing models, 2016

Qingnan Zhou and Alec Jacobson. Thingi10k: A dataset of 10,000 3d-printing models, 2016. 5, 6, 7, 2 Fast and Robust Mesh Simplification for Generated and Real-World 3D Assets Supplementary Material

work page 2016
[38]

virtual edge

Implementation Details This supplementary document provides additional imple- mentation and experimental details for FA-QEM, a feature- aware mesh simplification pipeline designed for modern 3D reconstruction and generative workflows. Our implementa- tion focuses on scalability, robustness, and efficiency, en- abling the conversion of dense, unstructured ...

work page
[39]

Hyperparameter Justification The performance of FA-QEM is governed by a small set of weighting parameters. Importantly, all hyperparameters used in the main paper are fixed across datasets and models, demonstrating that FA-QEM operates as a general-purpose method without requiring per-instance tuning. The values were determined empirically by testing on a...

work page
[40]

Its performance advantage over prior methods stems from both algorithmic design and implementation- level optimizations

Performance and Scalability FA-QEM is designed for efficient processing of large- scale 3D assets arising from reconstruction and generative pipelines. Its performance advantage over prior methods stems from both algorithmic design and implementation- level optimizations. Algorithmic Contributions to Performance:Our primary algorithmic performance gain co...

work page
[41]

thin-wall

Detailed Texture Mapping Validation We further evaluate the effectiveness of our texture transfer strategy, particularly in the context of modern 3D pipelines where high-quality appearance must be preserved after ag- gressive geometric simplification. 9.1. Quantitative Trade-off: Efficiency vs. Fidelity In the main paper, we noted that FA-QEM achieves tex...

work page
[42]

Robustness and Limitations This section provides a more detailed analysis of FA-QEM’s performance on the challenging cases, substantiating the ro- bustness claims made in the main paper. Large Texture-Varying Inputs:Our method is partic- ularly well-suited for models with complex texture layouts due to our philosophy of decoupling geometry from the UV par...

work page
[43]

In Figure 11, the first two meshes are from the Real World Textured Things [21] dataset and the last mesh is generated from Hunyuan3D model [33]

Additional Qualitative Results We present further qualitative results in the following fig- ures 11, 12, and 13. In Figure 11, the first two meshes are from the Real World Textured Things [21] dataset and the last mesh is generated from Hunyuan3D model [33]. This shows that our method is robust to simplify highly complex and non-manifold meshes. Similarly...

work page