pith. sign in

arxiv: 2603.27309 · v2 · pith:3ADWP3EGnew · submitted 2026-03-28 · 💻 cs.GR · cs.CV

MeshTailor: Cutting Seams via Generative Mesh Traversal

Xueqi Ma , Xingguang Yan , Congyue Zhang , Hui Huang This is my paper

Pith reviewed 2026-05-21 11:02 UTC · model grok-4.3

classification 💻 cs.GR cs.CV
keywords 3D mesh processingseam generationgenerative modelstransformerautoregressive predictiongraph traversalsurface cuttingmesh unfolding
0
0 comments X

The pith

MeshTailor generates coherent seams on 3D meshes by autoregressive traversal directly on the mesh graph.

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

The paper presents MeshTailor as a generative method that creates edge-aligned seams straight from the mesh structure rather than through external projections or optimization. It serializes the seam possibilities into ordered chains that start with large structural cuts and refine down to details, then feeds both connectivity and shape data into a transformer that predicts the next vertex along the seam one step at a time. This setup is intended to avoid the artifacts and manual fixes common in earlier approaches while producing seams that stay aligned with the surface edges. Readers interested in 3D modeling would care because cleaner seams simplify unfolding, texturing, and fabrication of complex shapes.

Core claim

MeshTailor is the first mesh-native generative framework for synthesizing edge-aligned seams on 3D surfaces. It operates directly on the mesh graph, eliminating projection artifacts and fragile snapping heuristics. ChainingSeams provides a hierarchical serialization of the seam graph that orders chains from global structural cuts down to local details in a coarse-to-fine manner, and a dual-stream encoder fuses topological and geometric context. The MeshTailor Transformer utilizes an autoregressive pointer layer to trace seams vertex-by-vertex within local neighborhoods, producing more coherent and structurally regular seam layouts than recent optimization-based and learning-based baselines.

What carries the argument

ChainingSeams hierarchical serialization of the seam graph combined with dual-stream vertex embeddings that feed an autoregressive pointer layer for vertex-by-vertex tracing inside local neighborhoods.

If this is right

  • Direct operation on the mesh graph removes dependence on extrinsic coordinates and associated projection errors.
  • Coarse-to-fine chaining lets the model build large structural seams first before adding fine details.
  • Autoregressive vertex prediction within neighborhoods produces edge-aligned seams without post-processing heuristics.
  • The resulting layouts show greater coherence and structural regularity than both optimization and earlier learning methods.

Where Pith is reading between the lines

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

  • The same traversal approach could be adapted for other mesh tasks such as generating cut paths for unfolding or procedural detailing.
  • If the hierarchical ordering generalizes, similar graph-native generators might reduce reliance on 2D projections across broader 3D geometry processing pipelines.
  • Applications like automated texture atlas creation or fabrication planning could incorporate this seam generator as a modular step.

Load-bearing premise

That ordering seams in global-to-local chains and embedding both topology and geometry together is enough for an autoregressive model to trace reliable seams without needing projections or snapping fixes.

What would settle it

A side-by-side quantitative comparison on a benchmark set of 3D meshes measuring seam coherence and structural regularity metrics to check whether MeshTailor outputs are measurably superior to optimization and prior learning baselines.

Figures

Figures reproduced from arXiv: 2603.27309 by Congyue Zhang, Hui Huang, Xingguang Yan, Xueqi Ma.

Figure 1
Figure 1. Figure 1: MeshTailor. Top: MeshTailor generates seams (colored lines) directly on 3D meshes, producing clean, semantically aligned cuts that respect natural shape structure. Bottom: The resulting seams partition surfaces into coherent UV charts, which are flattened into 2D layouts with minimal fragmentation. Right: These high-quality UV maps facilitate seamless texture application, as demonstrated by the final textu… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of MeshTailor. Left: The dual-stream encoder. The input mesh is processed in parallel: the top stream extracts topological connectivity features H via a Graph Encoder on the mesh topology {V, E}, while the bottom stream samples surface points to extract global shape semantics tokens Z using a pretrained point-cloud encoder (frozen during training). These representations are fused via cross-attenti… view at source ↗
Figure 3
Figure 3. Figure 3: Canonical ordering of seam chains (coarse-to-fine). We serialize an unordered seam set into a deterministic sequence for autoregressive training/inference with a loops-first, balance-first, large-patch-first strategy: we prioritize loop cuts over open chains, re￾peatedly select the largest remaining surface patch, and within that patch choose the loop cut that best balances the two resulting sub-patch area… view at source ↗
Figure 4
Figure 4. Figure 4: Step-by-step seam generation as mesh traversal. Our autoregressive decoder traverses the input mesh connectivity, se￾lecting the next vertex from the current 1-ring neighborhood. This local constraint makes the predicted seams mesh-native and avoids invalid jumps across the surface. 3. Method 3.1. Overview Our goal is to generate UV seams that are both edge-aligned and continuous (mesh-native by constructi… view at source ↗
Figure 5
Figure 5. Figure 5: Seam layout and area distortion comparison on GarmentCodeData [15]. For each method, we show the predicted seams on the 3D mesh (left) and the corresponding area distortion heatmap on the UV layout (right). While prior methods often produce fragmented or jagged cuts that lead to irregular UV islands, MeshTailor generates cleaner, garment-aligned seam structures with coherent chains and loops, resulting in … view at source ↗
Figure 6
Figure 6. Figure 6: Seam layout and UV area distortion comparison on TexVerse [56]. We qualitatively compare MeshTailor with OptCuts, xatlas, Nuvo, Blender Smart UV Project, and PartUV on diverse assets from TexVerse. Compared to existing baselines that often produce fragmented charts or jagged cut boundaries, MeshTailor generates cleaner, mesh-aligned seams with coherent long-range chain/loop struc￾tures while maintaining co… view at source ↗
Figure 7
Figure 7. Figure 7: Left: pairwise preference rates between methods, where [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: UV quality under a tiling stripe texture. Fragmented [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Production-oriented UV usability. (a) Our UV layout supports diverse texture appearances on the same garment. (b) Stripe tiling with a user-edited logo, showing results from prior methods. Tailor is consistently preferred over all baselines, validating that mesh-native seam chains better match production con￾ventions. See Supplemental Material C.2 for details. 4.4. Applications Texture editing and replacem… view at source ↗
Figure 11
Figure 11. Figure 11: MeshTailor generalizes to meshes at different levels of [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: We perturb the input mesh by adding Gaussian noise [PITH_FULL_IMAGE:figures/full_fig_p009_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Divide-and-conquer inference. Starting from a high-resolution mesh, MeshTailor predicts a major seam loop to split the surface into disconnected components, then recursively applies the same seam-tracing process to each sub-mesh. This recursion progressively decomposes the asset into semantically coherent parts, yielding clean, edge-aligned seams and the corresponding UV charts [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 14
Figure 14. Figure 14: Ablation on seam generation. We compare Mesh [PITH_FULL_IMAGE:figures/full_fig_p010_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Graph Encoder. We embed each vertex with coordinate-based point features and apply a stack of GraphSAGE layers with SiLU and LayerNorm to aggregate neighborhood in￾formation, producing per-vertex connectivity-aware embeddings used for seam generation. map a vertex id v ∈ [0, N−1] to candidate id (v + 2), while [EOC] and [EOS] are mapped to candidate ids 0 and 1, respectively. B.3.2. Encoding We compute th… view at source ↗
Figure 16
Figure 16. Figure 16: Coordinate-to-mesh projection. Coordinate-based seams can look plausible in Euclidean space, but snapping them to the input mesh often induces jagged boundaries and may collapse onto the wrong surface sheet under nearby/interpenetrating geom￾etry, leading to disconnections [PITH_FULL_IMAGE:figures/full_fig_p017_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: User study interface. We present participants with paired seam layout results (including seam visualizations on the 3D mesh and the corresponding UV maps) and ask them to choose the higher-quality option in a 2AFC setting, following the pro￾vided criteria (minimization, concealment, and geometry aware￾ness). Structural complexity. We report the number of charts (UV islands) as a proxy for cut complexity. … view at source ↗
Figure 18
Figure 18. Figure 18: Failure cases. (a) Hair and spiky assets. (b) Extremely low-poly meshes. (c) Decoding errors can derail a chain (red ar￾rows). Projection-induced jaggedness. Even when predicted segments form a visually plausible wireframe in Euclidean space, converting them into valid seams requires snapping points to mesh vertices/edges. Because the target is a dis￾crete mesh, small deviations in predicted coordinates c… view at source ↗
Figure 19
Figure 19. Figure 19: UV layout comparison on GarmentCodeData. Each island is shown with a unique color to reveal chart fragmentation and boundary regularity [PITH_FULL_IMAGE:figures/full_fig_p019_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: UV layout comparison on TexVerse. Each island is shown with a unique color to reveal chart fragmentation and boundary regularity [PITH_FULL_IMAGE:figures/full_fig_p020_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Additional qualitative results. MeshTailor produces coherent seam layouts across diverse categories and shapes. Colored curves denote different predicted seam chains overlaid on the input meshes [PITH_FULL_IMAGE:figures/full_fig_p021_21.png] view at source ↗
read the original abstract

We present MeshTailor, the first mesh-native generative framework for synthesizing edge-aligned seams on 3D surfaces. Unlike prior optimization-based or extrinsic learning-based methods, MeshTailor operates directly on the mesh graph, eliminating projection artifacts and fragile snapping heuristics. We introduce ChainingSeams, a hierarchical serialization of the seam graph that orders chains from global structural cuts down to local details in a coarse-to-fine manner, and a dual-stream encoder that fuses topological and geometric context. Leveraging this hierarchical representation and dual-stream vertex embeddings, our MeshTailor Transformer utilizes an autoregressive pointer layer to trace seams vertex-by-vertex within local neighborhoods. Extensive evaluations show that MeshTailor produces more coherent and structurally regular seam layouts compared to recent optimization-based and learning-based baselines.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript presents MeshTailor, the first mesh-native generative framework for synthesizing edge-aligned seams on 3D surfaces. It operates directly on the mesh graph using ChainingSeams, a hierarchical serialization of the seam graph that orders chains from global structural cuts to local details in a coarse-to-fine manner, together with a dual-stream encoder fusing topological and geometric context. The MeshTailor Transformer employs an autoregressive pointer layer to trace seams vertex-by-vertex within local neighborhoods. The central claim is that this approach yields more coherent and structurally regular seam layouts than recent optimization-based and learning-based baselines, supported by extensive evaluations.

Significance. If the central claims hold, the work offers a meaningful contribution to geometry processing by introducing a direct mesh-graph generative model that sidesteps projection artifacts and snapping heuristics common in prior extrinsic methods. The combination of hierarchical ChainingSeams serialization and dual-stream embeddings provides a novel mechanism for autoregressive seam tracing, which could influence future work on mesh segmentation, UV mapping, and related tasks. The paper's emphasis on native mesh operations is a clear strength, though the absence of detailed quantitative support in the abstract limits immediate evaluation of its practical impact.

major comments (2)
  1. Abstract: the claim that 'extensive evaluations show that MeshTailor produces more coherent and structurally regular seam layouts' is load-bearing for the superiority assertion, yet the abstract (and available summary) provides no quantitative metrics, baseline details, error analysis, or ablation results; this directly weakens the ability to verify the reported advantage over optimization and learning baselines.
  2. Method description of ChainingSeams and autoregressive pointer layer: the hierarchical serialization via ChainingSeams combined with dual-stream vertex embeddings is asserted to enable reliable autoregressive seam tracing without projection artifacts. For meshes with complex topology or long seams, however, sequential local vertex choices risk compounding prediction errors, potentially producing locally plausible but globally fragmented or irregular layouts; this assumption is central to the coherence claim and requires explicit discussion or additional global-consistency experiments.
minor comments (2)
  1. Abstract: the phrase 'mesh-native' is used without an immediate contrast to extrinsic methods; a brief parenthetical definition would improve clarity for readers unfamiliar with the distinction.
  2. Notation throughout: the dual-stream encoder is described as fusing 'topological and geometric context,' but the precise fusion mechanism (e.g., concatenation, attention, or gating) is not specified in the summary; adding an equation or diagram reference would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address each major comment below and have prepared revisions to strengthen the presentation of results and the discussion of model assumptions.

read point-by-point responses

  1. Referee: [—] Abstract: the claim that 'extensive evaluations show that MeshTailor produces more coherent and structurally regular seam layouts' is load-bearing for the superiority assertion, yet the abstract (and available summary) provides no quantitative metrics, baseline details, error analysis, or ablation results; this directly weakens the ability to verify the reported advantage over optimization and learning baselines.

    Authors: We agree that the abstract would benefit from concrete quantitative support for the central claim. In the revised manuscript we have updated the abstract to include a concise summary of the key metrics (coherence score, structural regularity index, and relative improvements versus the strongest baselines) drawn from the quantitative tables in Section 5. This keeps the abstract brief while making the superiority statement verifiable at a glance. revision: yes

  2. Referee: [—] Method description of ChainingSeams and autoregressive pointer layer: the hierarchical serialization via ChainingSeams combined with dual-stream vertex embeddings is asserted to enable reliable autoregressive seam tracing without projection artifacts. For meshes with complex topology or long seams, however, sequential local vertex choices risk compounding prediction errors, potentially producing locally plausible but globally fragmented or irregular layouts; this assumption is central to the coherence claim and requires explicit discussion or additional global-consistency experiments.

    Authors: We acknowledge the legitimate concern about error accumulation in long autoregressive sequences. The hierarchical ordering in ChainingSeams is intended to mitigate this by resolving global cuts before local refinements, and the dual-stream embeddings supply both local geometry and global topological context at each step. In the revision we have added an explicit paragraph in Section 3.3 discussing potential compounding errors and the mechanisms that limit their impact. We have also inserted a new global-consistency experiment (reported in Section 5.4) that measures end-to-end seam continuity on high-genus and long-seam meshes; the results show that fragmentation remains low relative to the baselines. revision: yes

Circularity Check

0 steps flagged

No significant circularity in MeshTailor derivation chain

full rationale

The paper introduces a new mesh-native generative approach with ChainingSeams for hierarchical seam serialization and a dual-stream encoder for topological-geometric fusion, followed by an autoregressive pointer-based tracing in the MeshTailor Transformer. These components are presented as novel constructions operating directly on the mesh graph, with evaluations against external optimization and learning baselines. No equations, predictions, or central claims reduce by construction to fitted inputs or self-citations; the framework is self-contained with independent content relative to prior work. No load-bearing uniqueness theorems or ansatzes from overlapping author citations are invoked in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The framework rests on the unstated premise that direct mesh-graph traversal can capture all necessary seam constraints without external projection; no free parameters, axioms, or invented entities are explicitly listed in the abstract.

pith-pipeline@v0.9.0 · 5662 in / 1063 out tokens · 34372 ms · 2026-05-21T11:02:52.490139+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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

59 extracted references · 59 canonical work pages · 2 internal anchors

  1. [1]

    Kim, Siddhartha Chaudhuri, Jun Saito, and Thibault Groueix

    Noam Aigerman, Kunal Gupta, Vladimir G. Kim, Siddhartha Chaudhuri, Jun Saito, and Thibault Groueix. Neural jaco- bian fields: learning intrinsic mappings of arbitrary meshes. ACM Trans. on Graphics (Proc. SIGGRAPH), 41(4):109:1– 109:17, 2022. 2

    work page 2022

  2. [2]

    Mesh simplification for unfolding

    Manas Bhargava, Camille Schreck, Marco Freire, Pierre- Alexandre Hugron, Sylvain Lefebvre, Silvia Sell ´an, and Bernd Bickel. Mesh simplification for unfolding. InCom- puter Graphics Forum, 2025. 3

    work page 2025

  3. [3]

    Blender manual: Seams from islands

    Blender. Blender manual: Seams from islands. https : / / docs . blender . org / manual / en / latest / modeling / meshes / uv / unwrapping / seams.html. 5, 13

    work page

  4. [4]

    Blender — a 3d modelling and rendering package.https://www.blender.org/

    Blender Online Community. Blender — a 3d modelling and rendering package.https://www.blender.org/. 3, 6

    work page

  5. [5]

    Kaufman, Qingnan Zhou, Shahar Z

    Xingyi Du, Danny M. Kaufman, Qingnan Zhou, Shahar Z. Kovalsky, Yajie Yan, Noam Aigerman, and Tao Ju. Optimiz- ing global injectivity for constrained parameterization.ACM Trans. on Graphics (Proc. SIGGRAPH Asia), 40(6):260:1– 260:18, 2021. 2

    work page 2021

  6. [6]

    Multiresolution analysis of arbitrary meshes

    Matthias Eck, Tony DeRose, Tom Duchamp, Hugues Hoppe, Michael Lounsbery, and Werner Stuetzle. Multiresolution analysis of arbitrary meshes. InProc. SIGGRAPH, page 173–182, 1995. 2

    work page 1995

  7. [7]

    Geome- try image diffusion: Fast and data-efficient text-to-3d with image-based surface representation

    Slava Elizarov, Ciara Rowles, and Simon Donn ´e. Geome- try image diffusion: Fast and data-efficient text-to-3d with image-based surface representation. InProc. Int. Conf. on Learning Representations, 2025. 3

    work page 2025

  8. [8]

    Michael S. Floater. Parametrization and smooth approxima- tion of surface triangulations.Comput. Aided Geom. Des., 14(3):231–250, 1997. 2

    work page 1997

  9. [9]

    Michael S. Floater. Mean value coordinates.Computer Aided Geometric Design, 20(1):19–27, 2003. 2

    work page 2003

  10. [10]

    Hamilton, Zhitao Ying, and Jure Leskovec

    William L. Hamilton, Zhitao Ying, and Jure Leskovec. In- ductive representation learning on large graphs. InProc. Conf. on Neural Information Processing Systems, pages 1024–1034, 2017. 3, 4

    work page 2017

  11. [11]

    Romero, Tsung-Yi Lin, and Ming-Yu Liu

    Zekun Hao, David W. Romero, Tsung-Yi Lin, and Ming-Yu Liu. Meshtron: High-fidelity, artist-like 3d mesh generation at scale, 2024. 3

    work page 2024

  12. [12]

    Mesh pa- rameterization: theory and practice

    Kai Hormann, Konrad Polthier, and Alia Sheffer. Mesh pa- rameterization: theory and practice. InACM SIGGRAPH ASIA 2008 Courses, 2008. 1, 2

    work page 2008

  13. [13]

    xatlas: Mesh uv unwrapping library

    jpcy. xatlas: Mesh uv unwrapping library. GitHub repository,

    work page

  14. [14]

    D-charts: Quasi-developable mesh segmentation.Computer Graphics Forum, 24(3), 2005

    Dan Julius, Vladislav Kraevoy, and Alla Sheffer. D-charts: Quasi-developable mesh segmentation.Computer Graphics Forum, 24(3), 2005. 3

    work page 2005

  15. [15]

    Garmentcodedata: A dataset of 3d made-to-measure garments with sewing pat- terns

    Maria Korosteleva, Timur Levent Kesdogan, Fabian Kem- per, Stephan Wenninger, Jasmin Koller, Yuhan Zhang, Mario Botsch, and Olga Sorkine-Hornung. Garmentcodedata: A dataset of 3d made-to-measure garments with sewing pat- terns. InProc. Euro. Conf. on Computer Vision, pages 110– 127, 2024. 5, 6, 13

    work page 2024

  16. [16]

    Least squares conformal maps for automatic tex- ture atlas generation.ACM Trans

    Bruno L ´evy, Sylvain Petitjean, Nicolas Ray, and J ´erome Maillot. Least squares conformal maps for automatic tex- ture atlas generation.ACM Trans. on Graphics, 21(3):362:1– 362:10, 2002. 2

    work page 2002

  17. [17]

    Kaufman, Vladimir G

    Minchen Li, Danny M. Kaufman, Vladimir G. Kim, Justin Solomon, and Alla Sheffer. Optcuts: joint optimization of surface cuts and parameterization.ACM Trans. on Graphics (Proc. SIGGRAPH Asia), 37(6):247:1–247:13, 2018. 1, 2, 6

    work page 2018

  18. [18]

    Garma- genet: A multimodal generative framework for sewing pat- tern design and generic garment modeling.ACM Transac- tions on Graphics (TOG), 44(6):1–23, 2025

    Siran Li, Ruiyang Liu, Chen Liu, Zhendong Wang, Gaofeng He, Yong-Lu Li, Xiaogang Jin, and Huamin Wang. Garma- genet: A multimodal generative framework for sewing pat- tern design and generic garment modeling.ACM Transac- tions on Graphics (TOG), 44(6):1–23, 2025. 3

    work page 2025

  19. [19]

    Auto-regressive surface cutting, 2025

    Yang Li, Victor Cheung, Xinhai Liu, Yuguang Chen, Zhongjin Luo, Biwen Lei, Haohan Weng, Zibo Zhao, Jing- wei Huang, Zhuo Chen, and Chunchao Guo. Auto-regressive surface cutting, 2025. 2, 3, 9

    work page 2025

  20. [20]

    PARTFIELD: learning 3d feature fields for part segmentation and beyond

    Ming-Yu Liu, Mikaela Angelina Uy, Donglai Xiang, Hao Su, Sanja Fidler, Nicholas Sharp, and Jun Gao. PARTFIELD: learning 3d feature fields for part segmentation and beyond

    work page

  21. [21]

    Kim, and Rana Hanocka

    Richard Liu, Noam Aigerman, Vladimir G. Kim, and Rana Hanocka. DA wand: Distortion-aware selection using neural mesh parameterization. InProc. IEEE/CVF Conf. on Com- puter Vision & Pattern Recognition, pages 16739–16749,

    work page

  22. [22]

    Seamless: seam erasure and seam-aware decou- pling of shape from mesh resolution.ACM Trans

    Songrun Liu, Zachary Ferguson, Alec Jacobson, and Yotam Gingold. Seamless: seam erasure and seam-aware decou- pling of shape from mesh resolution.ACM Trans. on Graph- ics (Proc. SIGGRAPH Asia), 36(6):216:1–216:15, 2017. 13

    work page 2017

  23. [23]

    Generating 3d house wireframes with semantics

    Xueqi Ma, Yilin Liu, Wenjun Zhou, Ruowei Wang, and Hui Huang. Generating 3d house wireframes with semantics. In ECCV, pages 223–240, 2024. 3

    work page 2024

  24. [24]

    Clr-wire: Towards continuous latent representations for 3d curve wireframe generation

    Xueqi Ma, Yilin Liu, Tianlong Gao, Qirui Huang, and Hui Huang. Clr-wire: Towards continuous latent representations for 3d curve wireframe generation. InACM SIGGRAPH, pages 77:1–77:11, 2025. 3

    work page 2025

  25. [25]

    Autosdf: Shape priors for 3d comple- tion, reconstruction and generation

    Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shub- ham Tulsiani. Autosdf: Shape priors for 3d comple- tion, reconstruction and generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 306–315, 2022. 3

    work page 2022

  26. [26]

    Kim, and Niloy J

    Luca Morreale, Noam Aigerman, Vladimir G. Kim, and Niloy J. Mitra. Neural surface maps. InProc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition, pages 4639–4648, 2021. 2

    work page 2021

  27. [27]

    Spectral conformal parameterization

    Patrick Mullen, Yiying Tong, Pierre Alliez, and Mathieu Desbrun. Spectral conformal parameterization. InProc. Eu- rographics Symp. on Geometry Processing, pages 1487:1– –1487:8, 2008. 2

    work page 2008

  28. [28]

    Charlie Nash, Yaroslav Ganin, S. M. Ali Eslami, and Pe- ter W. Battaglia. Polygen: An autoregressive generative model of 3d meshes. InProc. Int. Conf. on Machine Learn- ing, pages 7220–7229. PMLR, 2020. 3

    work page 2020

  29. [29]

    Autocuts: simultaneous dis- tortion and cut optimization for uv mapping.ACM Trans

    Roi Poranne, Marco Tarini, Sandro Huber, Daniele Panozzo, and Olga Sorkine-Hornung. Autocuts: simultaneous dis- tortion and cut optimization for uv mapping.ACM Trans. on Graphics (Proc. SIGGRAPH Asia), 36(6):215:1–215:11,

    work page

  30. [30]

    Scalable locally injective mappings

    Michael Rabinovich, Roi Poranne, Daniele Panozzo, and Olga Sorkine-Hornung. Scalable locally injective mappings. ACM Trans. on Graphics, 36(2):16:1–16:16, 2017. 2

    work page 2017

  31. [31]

    Sander, John Snyder, Steven J

    Pedro V . Sander, John Snyder, Steven J. Gortler, and Hugues Hoppe. Texture mapping progressive meshes. InProc. SIG- GRAPH, pages 409:1–416:8, 2001. 3

    work page 2001

  32. [32]

    Boundary first flatten- ing.ACM Trans

    Rohan Sawhney and Keenan Crane. Boundary first flatten- ing.ACM Trans. on Graphics, 37(1):5:1–5:14, 2017. 2

    work page 2017

  33. [33]

    Variational surface cut- ting.ACM Trans

    Nicholas Sharp and Keenan Crane. Variational surface cut- ting.ACM Trans. on Graphics (Proc. SIGGRAPH), 37(4): 156:1–156:13, 2018. 2

    work page 2018

  34. [34]

    Surface parameterization for meshing by triangulation flattening, 2000

    Alla Sheffer and Eric De Sturler. Surface parameterization for meshing by triangulation flattening, 2000. 2

    work page 2000

  35. [35]

    Abf++: fast and robust angle based flat- tening.ACM Trans

    Alla Sheffer, Bruno L ´evy, Maxim Mogilnitsky, and Alexan- der Bogomyakov. Abf++: fast and robust angle based flat- tening.ACM Trans. on Graphics, 24(2):311:1–311:20, 2005. 2, 5

    work page 2005

  36. [36]

    Mesh param- eterization methods and their applications.Found

    Alla Sheffer, Emil Praun, and Kenneth Rose. Mesh param- eterization methods and their applications.Found. Trends. Comput. Graph. Vis., 2(2):105–171, 2006. 1

    work page 2006

  37. [37]

    Meshgpt: Generating triangle meshes with decoder-only transformers

    Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, Tatiana Tommasi, Daniele Sirigatti, Vladislav Rosov, Angela Dai, and Matthias Nießner. Meshgpt: Generating triangle meshes with decoder-only transformers. InProc. IEEE/CVF Conf. on Computer Vision & Pattern Recognition, pages 19615– 19625, 2024. 3

    work page 2024

  38. [38]

    Sorkine, D

    O. Sorkine, D. Cohen-Or, R. Goldenthal, and D. Lischin- ski. Bounded-distortion piecewise mesh parameterization. In IEEE Visualization, 2002. VIS 2002., pages 355–362, 2002. 2

    work page 2002

  39. [39]

    Srinivasan, Stephan J

    Pratul P. Srinivasan, Stephan J. Garbin, Dor Verbin, Jonathan T. Barron, and Ben Mildenhall. Nuvo: Neural uv mapping for unruly 3d representations. InProc. Euro. Conf. on Computer Vision, pages 18–34, 2024. 1, 2, 6

    work page 2024

  40. [40]

    Roformer: Enhanced transformer with rotary position embedding, 2023

    Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding, 2023. 15

    work page 2023

  41. [41]

    Qwen3-VL Technical Report

    Qwen Team. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631, 2025. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  42. [42]

    Graphseam: Super- vised graph learning framework for semantic uv mapping,

    Fatemeh Teimury, Bruno Roy, Juan Sebasti ´an Casallas, David MacDonald, and Mark Coates. Graphseam: Super- vised graph learning framework for semantic uv mapping,

    work page

  43. [43]

    How to draw a graph.Proceedings of the London Mathematical Society, 3(1):743–767, 1963

    William Thomas Tutte. How to draw a graph.Proceedings of the London Mathematical Society, 3(1):743–767, 1963. 2

    work page 1963

  44. [44]

    Pointer networks

    Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. Pointer networks. InProc. Conf. on Neural Information Processing Systems, page 2692–2700, 2015. 2

    work page 2015

  45. [45]

    LLaMA-Mesh: Unifying 3d mesh generation with language models.arXiv preprint arXiv:2411.09595, 2024

    Zhengyi Wang, Jonathan Lorraine, Yikai Wang, Hang Su, Jun Zhu, Sanja Fidler, and Xiaohui Zeng. Llama-mesh: Unifying 3d mesh generation with language models.arXiv preprint arXiv:2411.09595, 2024. 3

    work page arXiv 2024

  46. [46]

    Partuv: Part-based uv unwrap- ping of 3d meshes

    Zhaoning Wang, Xinyue Wei, Ruoxi Shi, Xiaoshuai Zhang, Hao Su, and Minghua Liu. Partuv: Part-based uv unwrap- ping of 3d meshes. InProc. SIGGRAPH Asia, pages 15:1– 15:12, 2025. 2, 3, 5, 6, 16

    work page 2025

  47. [47]

    Octgpt: Octree-based multi- scale autoregressive models for 3d shape generation

    Si-Tong Wei, Rui-Huan Wang, Chuan-Zhi Zhou, Baoquan Chen, and Peng-Shuai Wang. Octgpt: Octree-based multi- scale autoregressive models for 3d shape generation. InPro- ceedings of the Special Interest Group on Computer Graph- ics and Interactive Techniques Conference Conference Pa- pers, pages 1–11, 2025. 3

    work page 2025

  48. [48]

    Structured 3D Latents for Scalable and Versatile 3D Generation

    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.arXiv preprint arXiv:2412.01506, 2024. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  49. [49]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tomp- kin, Vincent Sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. InComputer graphics forum, pages 641–676. Wiley Online Library, 2022. 3

    work page 2022

  50. [50]

    Seam- crafter: Enhancing mesh seam generation for artist uv un- wrapping via reinforcement learning, 2025

    Duoteng Xu, Yuguang Chen, Jing Li, Xinhai Liu, Xueqi Ma, Zhuo Chen, Dongyu Zhang, and Chunchao Guo. Seam- crafter: Enhancing mesh seam generation for artist uv un- wrapping via reinforcement learning, 2025. 3

    work page 2025

  51. [51]

    Mitra, Dani Lischin- ski, Daniel Cohen-Or, and Hui Huang

    Xingguang Yan, Liqiang Lin, Niloy J. Mitra, Dani Lischin- ski, Daniel Cohen-Or, and Hui Huang. Shapeformer: Transformer-based shape completion via sparse representa- tion. InProc. IEEE/CVF Conf. on Computer Vision & Pat- tern Recognition, pages 6229–6239, 2022. 3

    work page 2022

  52. [52]

    Xingguang Yan, Han-Hung Lee, Ziyu Wan, and Angel X. Chang. An object is worth 64x64 pixels: Generating 3d ob- ject via image diffusion, 2024. 3

    work page 2024

  53. [53]

    Sampart3d: Segment any part in 3d objects.arXiv preprint arXiv:2411.07184, 2024

    Yunhan Yang, Yukun Huang, Yuan-Chen Guo, Liangjun Lu, Xiaoyang Wu, Edmund Y Lam, Yan-Pei Cao, and Xihui Liu. Sampart3d: Segment any part in 3d objects.arXiv preprint arXiv:2411.07184, 2024. 3

    work page arXiv 2024

  54. [54]

    3dshape2vecset: A 3d shape representation for neu- ral fields and generative diffusion models.ACM Trans

    Biao Zhang, Jiapeng Tang, Matthias Nießner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neu- ral fields and generative diffusion models.ACM Trans. on Graphics (Proc. SIGGRAPH), 42(4):92:1–92:16, 2023. 3

    work page 2023

  55. [55]

    Flatten anything: unsupervised neural surface parameteriza- tion

    Qijian Zhang, Junhui Hou, Wenping Wang, and Ying He. Flatten anything: unsupervised neural surface parameteriza- tion. InProc. Conf. on Neural Information Processing Sys- tems, pages 93–113, 2024. 1, 2

    work page 2024

  56. [56]

    Texverse: A universe of 3d objects with high-resolution textures, 2025

    Yibo Zhang, Li Zhang, Rui Ma, and Nan Cao. Texverse: A universe of 3d objects with high-resolution textures, 2025. 5, 7, 13

    work page 2025

  57. [57]

    Michelangelo: conditional 3d shape generation based on shape-image-text aligned latent representation

    Zibo Zhao, Wen Liu, Xin Chen, Xianfang Zeng, Rui Wang, Pei Cheng, Bin Fu, Tao Chen, Gang Yu, and Shenghua Gao. Michelangelo: conditional 3d shape generation based on shape-image-text aligned latent representation. InProc. Conf. on Neural Information Processing Systems, 2023. 5, 15

    work page 2023

  58. [58]

    Iso-charts: stretch-driven mesh parameterization us- ing spectral analysis

    Kun Zhou, John Synder, Baining Guo, and Heung-Yeung Shum. Iso-charts: stretch-driven mesh parameterization us- ing spectral analysis. InProc. Eurographics Symp. on Ge- ometry Processing, pages 45–54, 2004. 3 MeshTailor: Cutting Seams via Generative Mesh Traversal Supplemental Material Xueqi Ma1 Xingguang Yan2 Congyue Zhang1 Hui Huang1* 1Shenzhen Universit...

    work page 2004

  59. [59]

    In addition, we introduce a chain-local positional embedding that resets after every [EOC]to encode within-chain progress

    to encode token order. In addition, we introduce a chain-local positional embedding that resets after every [EOC]to encode within-chain progress. Letπ t denote the number of steps since the most recent[EOC](starting from 0at each chain start). We add ˜et ←e t +PosEmb chain πt , which helps the model distinguish early versus late vertices inside the curren...

    work page