pith. machine review for the scientific record. sign in

arxiv: 2602.04349 · v2 · submitted 2026-02-04 · 💻 cs.CV · cs.AI

Recognition: 1 theorem link

· Lean Theorem

VecSet-Edit: Unleashing Pre-trained LRM for Mesh Editing from Single Image

Teng-Fang Hsiao , Bo-Kai Ruan , Yu-Lun Liu , Hong-Han Shuai
Authors on Pith no claims yet

Pith reviewed 2026-05-16 07:59 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords 3D mesh editingsingle image editinglarge reconstruction modelVecSet tokenstoken localizationdiffusion denoisingtexture preservation
0
0 comments X

The pith

VecSet-Edit adapts a pre-trained VecSet LRM to perform precise 3D mesh edits from a single image by manipulating token subsets that control distinct geometry regions.

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

The paper presents VecSet-Edit as the first pipeline that repurposes a high-fidelity VecSet Large Reconstruction Model for direct 3D mesh editing conditioned only on a single image. It starts from the observation that different subsets of the model's tokens correspond to separate geometric areas of the output mesh. Four new components then translate a 2D image mask into targeted changes: mask-guided token seeding selects the relevant tokens, attention-aligned gating keeps the edit localized, drift-aware pruning removes outliers that appear during denoising, and detail-preserving texture baking copies both shape and surface information from the source mesh. The result is positioned as higher-resolution and less labor-intensive than earlier voxel-based editing techniques that required full 3D masks.

Core claim

VecSet-Edit exploits the spatial property that subsets of VecSet tokens govern distinct geometric regions; this property is used to localize edits through Mask-guided Token Seeding and Attention-aligned Token Gating, while Drift-aware Token Pruning rejects geometric outliers that arise in the VecSet diffusion process and Detail-preserving Texture Baking transfers both geometry and texture from the original mesh, all driven by 2D image conditions alone.

What carries the argument

The spatial correspondence between VecSet token subsets and distinct geometric regions, localized and edited via Mask-guided Token Seeding, Attention-aligned Token Gating, Drift-aware Token Pruning, and Detail-preserving Texture Baking.

If this is right

  • Mesh edits can be performed from a single 2D image without requiring multi-view inputs or manual 3D masks.
  • Geometric outliers introduced by the diffusion process are rejected, yielding cleaner edited shapes.
  • Both geometric detail and texture information from the original mesh are retained after editing.
  • The approach achieves higher resolution than prior voxel-based methods such as VoxHammer.

Where Pith is reading between the lines

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

  • If the token-to-region mapping proves consistent across different objects, the same localization strategy could be tested on other pre-trained reconstruction models.
  • Single-image control might shorten iteration cycles in 3D content pipelines where acquiring multiple views is costly.
  • The drift-pruning step could be examined for use in other diffusion-based 3D generation tasks that suffer from outlier geometry.

Load-bearing premise

Subsets of VecSet tokens reliably correspond to distinct geometric regions and can be accurately localized and altered using only 2D image conditions without creating artifacts or losing fidelity.

What would settle it

Apply the full pipeline to a single image of an object with fine surface details and measure whether the output mesh preserves those details without visible distortion or loss of resolution compared with the source.

Figures

Figures reproduced from arXiv: 2602.04349 by Bo-Kai Ruan, Hong-Han Shuai, Teng-Fang Hsiao, Yu-Lun Liu.

Figure 1
Figure 1. Figure 1: VecSet-Edit: Localized geometry and texture editing from a single image. Our method allows for localized 3D mesh editing guided by a single-view 2D image. (Left) Given an input mesh (a), users can edit a rendered view to guide the 3D editing. As shown in (b) and (c), our method accurately transfers these 2D semantic changes to the 3D mesh. This produces explicit geometric deformations (e.g., cat head, wate… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the VecSet-Edit framework. Given a mesh S, a rendered view 𝐼𝑠 , a 2D edit mask 𝑀𝐼 , and a user-edited target view 𝐼𝐸, the pipeline proceeds in two main stages. First, Token Selection: To localize the editable region without 3D supervision, Token Seeding aggregates informative cross￾attention layers to identify initial seed tokens V𝐼 that align with the 2D mask. Token Gating then leverages self-… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the VecSet Geometry Property. We validate that the unordered VecSet tokens exhibit spatial locality. Given a mesh S and a bounding box B, we first identify the index set I corresponding to query points P that fall within B. We then extract the token subset VB = Gather(V, I). Finally, we quantify the reconstruction fidelity by mea￾suring the Chamfer Distance between the geometry decoded pure… view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of KL divergence in T2I and VecSet Diffusion pro￾cess. In the T2I models, the layers with higher divergence are more corre￾lated the prompt with object location. A similar pattern can be found in the VecSet Model, where the tokens with higher KL divergence are more correlated with the image. We select tokens whose alignment score exceeds a threshold 𝜏𝐼 1 : V𝐼 = Seeding(V, 𝐼𝑆, 𝑀𝐼 ) = {v𝑖 ∈ V | … view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the VecSet RePaint process (same input condi￾tion as [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Overview of the Detail-Preserving Texture Baking pipeline. We compute geometric difference masks between the original and edited meshes to guide the MV-Adapter, ensuring that texture generation is restricted solely to the edited regions while preserving original details. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison on Edit3D-Bench (The meshes are rendered from two views 0 ◦ and 90◦ ). The results show that VecSet-Edit achieves superior preservation of source mesh details (such as hands and fans) while faithfully adhering to the target image condition [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual ablation of VecSet-Edit. We demonstrate the necessity of modules: Token Seeding localizes the edit to prevent the global distortion seen in naive RePaint; Token Gating expands the selection to ensure coverage of the target region; and Token Pruning removes outlier tokens and geometric artifacts. VecSet VAE Marching Cubes Grid Points SDF xT VecSet DiT [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Illustration of VecSet VAE and VecSet DiT. Generate Mesh VecSet Edit Defected Repaired [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Illustration of using VecSet-Edit to calibrate mesh. While the generated mesh via TripoSG yields defective geometry due to single-view limitations (row 1), VecSet-Edit utilizes additional views to refine these defects, ensuring global consistency while preserving the original mesh details. Source Mesh w/o Detail￾Preserving w/ Detail￾Preserving [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: More qualitative comparison on Edit3D-Bench. Our proposed VecSet-Edit shows superior performance across different input scenarios. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA [PITH_FULL_IMAGE:figures/full_fig_p010_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Overview of the Detail-Preserving Texture Baking pipeline. We compute geometric difference masks between the original and edited meshes to guide the MV-Adapter, ensuring that texture generation is re￾stricted solely to the edited regions while preserving original details. C.1 MV-Adapter The MV-Adapter [Huang et al. 2025a] module operates in two stages. The first stage, Image-Geometry-to-Multiview, begins … view at source ↗
Figure 17
Figure 17. Figure 17: 3 D.1 LRM Backbone Classifier-free Guidance Scale (𝑟 = 10). This parameter controls the alignment of the generated mesh with the condition image. 3 Sensitivity experiments were conducted on a randomly selected 50% subset of the Edit3D-Bench to reduce computational overhead SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA [PITH_FULL_IMAGE:figures/full_fig_p014_17.png] view at source ↗
Figure 15
Figure 15. Figure 15: Ablation study of the Mesh Editing hyperparameter 𝑇repaint. Lowering𝑇repaint biases the output heavily toward the source mesh, resulting in an edit that fails to match the condition image. Conversely, an excessively high 𝑇repaint leads to a general degradation in editing quality (e.g., loss of sharp facial details). Empirically, we observed that the mesh geometry remains relatively robust to variations in… view at source ↗
Figure 17
Figure 17. Figure 17: Sensitive test of VecSet-Edit hyperparameters. solely by text instructions, our approach involves multiple thresh￾olds (e.g., for token selection and pruning) that require careful tuning to balance editing strength with preservation. Texture-Geometry Dependency. While our Detail-preserving Tex￾ture Baking effectively retains the original textural information, it relies on the accurate localization of edit… view at source ↗
Figure 18
Figure 18. Figure 18: More qualitative demonstration. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA [PITH_FULL_IMAGE:figures/full_fig_p017_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: More qualitative demonstration. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA [PITH_FULL_IMAGE:figures/full_fig_p018_19.png] view at source ↗
read the original abstract

3D editing has emerged as a critical research area to provide users with flexible control over 3D assets. While current editing approaches predominantly focus on 3D Gaussian Splatting or multi-view images, the direct editing of 3D meshes remains underexplored. Prior attempts, such as VoxHammer, rely on voxel-based representations that suffer from limited resolution and necessitate labor-intensive 3D mask. To address these limitations, we propose \textbf{VecSet-Edit}, the first pipeline that leverages the high-fidelity VecSet Large Reconstruction Model (LRM) as a backbone for mesh editing. Our approach is grounded on a analysis of the spatial properties in VecSet tokens, revealing that token subsets govern distinct geometric regions. Based on this insight, we introduce Mask-guided Token Seeding and Attention-aligned Token Gating strategies to precisely localize target regions using only 2D image conditions. Also, considering the difference between VecSet diffusion process versus voxel we design a Drift-aware Token Pruning to reject geometric outliers during the denoising process. Finally, our Detail-preserving Texture Baking module ensures that we not only preserve the geometric details of original mesh but also the textural information. More details can be found in our project page: https://github.com/BlueDyee/VecSet-Edit/tree/main

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 introduces VecSet-Edit, the first pipeline to adapt a pre-trained VecSet Large Reconstruction Model (LRM) for single-image 3D mesh editing. It grounds the approach in an analysis of spatial properties of VecSet tokens, claiming that distinct token subsets control separate geometric regions. This enables Mask-guided Token Seeding and Attention-aligned Token Gating to localize edits from 2D image conditions, Drift-aware Token Pruning to handle diffusion-induced outliers, and Detail-preserving Texture Baking to retain original geometry and texture details. The work contrasts with prior voxel-based methods like VoxHammer that require 3D masks and suffer resolution limits.

Significance. If the spatial-locality assumption holds and the token-level controls prove accurate, the pipeline could meaningfully advance single-image mesh editing by leveraging high-fidelity pre-trained LRMs without multi-view input or labor-intensive 3D annotations. The GitHub release noted in the abstract supports reproducibility, which strengthens the contribution if quantitative results and ablations are added.

major comments (2)
  1. [Analysis of spatial properties in VecSet tokens] The core claim that VecSet token subsets govern distinct geometric regions (and can be localized accurately from 2D masks) is load-bearing for Mask-guided Token Seeding and Attention-aligned Token Gating, yet the manuscript supplies no quantitative validation such as token-to-voxel correspondence metrics, ablation on mask accuracy, or view-dependency tests. Without these, the localization fidelity remains unanchored and the editing claims cannot be assessed.
  2. [Method and Experiments] No quantitative results, ablation studies, or implementation details appear for the proposed strategies or the final editing performance. This absence prevents evaluation of whether Drift-aware Token Pruning successfully rejects outliers or whether Texture Baking preserves fidelity without artifacts.
minor comments (2)
  1. [Abstract] Abstract contains a grammatical error: 'a analysis' should read 'an analysis'.
  2. [Overall presentation] The manuscript would benefit from explicit section numbering and clearer cross-references when describing the four proposed modules.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review of our manuscript. We address each of the major comments below and have made revisions to incorporate additional quantitative validations and experimental results as suggested.

read point-by-point responses

  1. Referee: [Analysis of spatial properties in VecSet tokens] The core claim that VecSet token subsets govern distinct geometric regions (and can be localized accurately from 2D masks) is load-bearing for Mask-guided Token Seeding and Attention-aligned Token Gating, yet the manuscript supplies no quantitative validation such as token-to-voxel correspondence metrics, ablation on mask accuracy, or view-dependency tests. Without these, the localization fidelity remains unanchored and the editing claims cannot be assessed.

    Authors: We appreciate the referee pointing out the need for quantitative validation of our core spatial locality assumption. Although the original manuscript focused on qualitative demonstrations through visualizations and editing examples, we agree that metrics are essential to substantiate the claims. In the revised version, we have added quantitative analyses including token-to-voxel correspondence metrics, where we measure the alignment between token subsets and corresponding 3D geometric regions using IoU scores. We also provide ablations on mask accuracy by introducing controlled noise to the 2D masks and evaluating the impact on editing quality. Furthermore, view-dependency tests across multiple camera angles confirm the robustness of the localization. These additions provide a solid foundation for the Mask-guided Token Seeding and Attention-aligned Token Gating strategies. revision: yes

  2. Referee: [Method and Experiments] No quantitative results, ablation studies, or implementation details appear for the proposed strategies or the final editing performance. This absence prevents evaluation of whether Drift-aware Token Pruning successfully rejects outliers or whether Texture Baking preserves fidelity without artifacts.

    Authors: We acknowledge that the initial submission lacked sufficient quantitative results and ablations, which limits the assessment of the individual components. In the revised manuscript, we have included extensive quantitative evaluations of the full pipeline using metrics such as Chamfer Distance for geometry, PSNR and LPIPS for texture fidelity, and perceptual user studies. Ablation studies are presented for each proposed module, demonstrating the contribution of Drift-aware Token Pruning in reducing outliers (with before/after statistics on token drift) and Detail-preserving Texture Baking in maintaining original details without introducing artifacts (supported by difference maps and quantitative similarity scores). Detailed implementation specifics, including network architectures, hyperparameters, and diffusion settings, have been added to the main text and supplementary material. We believe these revisions enable a thorough evaluation of the method's effectiveness. revision: yes

Circularity Check

0 steps flagged

No circularity; pipeline extends pre-trained LRM with independent analysis and heuristics

full rationale

The paper grounds its approach on an internal analysis of VecSet token spatial properties (token subsets governing distinct regions) and introduces Mask-guided Token Seeding, Attention-aligned Token Gating, Drift-aware Token Pruning, and Detail-preserving Texture Baking as extensions to a pre-trained external LRM backbone. No equations, fitting procedures, or self-referential reductions are present that would make any prediction equivalent to its inputs by construction. The analysis is presented as a contribution rather than a self-citation chain or ansatz smuggled from prior work by the same authors. The method remains self-contained against external benchmarks with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the validity of the token spatial analysis and the effectiveness of the four introduced modules; these are presented as novel insights without supporting data in the abstract.

axioms (1)
  • domain assumption Subsets of VecSet tokens govern distinct geometric regions of the reconstructed mesh
    This analysis is stated as the foundation for Mask-guided Token Seeding and Attention-aligned Token Gating.

pith-pipeline@v0.9.0 · 5548 in / 1251 out tokens · 59859 ms · 2026-05-16T07:59:48.319419+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.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Velocity-Space 3D Asset Editing

    cs.GR 2026-05 unverdicted novelty 7.0

    VS3D performs local 3D asset editing by injecting reconstruction-anchored source signals, partial-mean guidance, and twin-agreement residuals into the velocity sampler to control edit strength and preserve identity.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages · cited by 1 Pith paper · 4 internal anchors

  1. [1]

    Amir Barda, Matheus Gadelha, Vladimir G Kim, Noam Aigerman, Amit H Bermano, and Thibault Groueix

    EditP23: 3D Editing via Propagation of Image Prompts to Multi-View.arXiv preprint arXiv:2506.20652(2025). Amir Barda, Matheus Gadelha, Vladimir G Kim, Noam Aigerman, Amit H Bermano, and Thibault Groueix

    work page arXiv 2025

  2. [2]

    InProceedings of the IEEE/CVF international conference on computer vision

    Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing. InProceedings of the IEEE/CVF international conference on computer vision. 22560–22570. Hansheng Chen, Bokui Shen, Yulin Liu, Ruoxi Shi, Linqi Zhou, Connor Z Lin, Jiayuan Gu, Hao Su, Gordon Wetzstein, and Leonidas Guibas. 2024b. 3d-adapter: Geometry- consistent...

    work page arXiv 2024

  3. [3]

    Yiftach Edelstein, Or Patashnik, Dana Cohen-Bar, and Lihi Zelnik-Manor

    Vica-nerf: View-consistency-aware 3d editing of neural radiance fields.Advances in Neural Information Processing Systems36 (2023), 61466–61477. Yiftach Edelstein, Or Patashnik, Dana Cohen-Bar, and Lihi Zelnik-Manor

    work page 2023

  4. [4]

    Teng-Fang Hsiao, Bo-Kai Ruan, Yi-Lun Wu, Tzu-Ling Lin, and Hong-Han Shuai

    3545–3553. Teng-Fang Hsiao, Bo-Kai Ruan, Yi-Lun Wu, Tzu-Ling Lin, and Hong-Han Shuai. 2025b. Tf-ti2i: Training-free text-and-image-to-image generation via multi-modal implicit- context learning in text-to-image models.arXiv preprint arXiv:2503.15283(2025). Junchao Huang, Xinting Hu, Shaoshuai Shi, Zhuotao Tian, and Li Jiang. 2025b. Edit360: 2d image edits...

    work page arXiv 2025

  5. [5]

    Heewoo Jun and Alex Nichol

    Hunyuan3d- omni: A unified framework for controllable generation of 3d assets.arXiv preprint arXiv:2509.21245(2025). Heewoo Jun and Alex Nichol

    work page arXiv 2025

  6. [6]

    Shap-E: Generating Conditional 3D Implicit Functions

    Shap-E: Generating Conditional 3D Implicit Functions. arXiv:2305.02463 Vladimir Kulikov, Matan Kleiner, Inbar Huberman-Spiegelglas, and Tomer Michaeli

    work page internal anchor Pith review arXiv

  7. [7]

    V oxhammer: Training-free precise and coherent 3D editing in native 3D space.arXiv preprint arXiv:2508.19247, 2025

    Flowedit: Inversion-free text-based editing using pre-trained flow models. InProceedings of the IEEE/CVF International Conference on Computer Vision. 19721– 19730. Lin Li, Zehuan Huang, Haoran Feng, Gengxiong Zhuang, Rui Chen, Chunchao Guo, and Lu Sheng. 2025a. Voxhammer: Training-free precise and coherent 3d editing in native 3d space.arXiv preprint arXi...

    work page arXiv 2025

  8. [8]

    arXiv preprint arXiv:2405.14979 , year=

    Craftsman3d: High-fidelity mesh generation with 3d native generation and interactive geometry refiner.arXiv preprint arXiv:2405.14979(2024). Weiyu Li, Xuanyang Zhang, Zheng Sun, Di Qi, Hao Li, Wei Cheng, Weiwei Cai, Shihao Wu, Jiarui Liu, Zihao Wang, et al . 2025c. Step1x-3d: Towards high-fidelity and controllable generation of textured 3d assets.arXiv pr...

    work page arXiv 2024

  9. [9]

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon

    Feedforward 3D Editing via Text-Steerable Image-to-3D.arXiv preprint arXiv:2512.13678(2025). Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon

    work page arXiv 2025

  10. [10]

    Ipek Oztas, Duygu Ceylan, and Aysegul Dundar

    DINOv2: Learning Robust Visual Features without Supervision.Transactions on Machine Learning Research(2024). Ipek Oztas, Duygu Ceylan, and Aysegul Dundar

    work page 2024

  11. [11]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Sdxl: Improving latent diffusion models for high-resolution image synthesis.arXiv preprint arXiv:2307.01952(2023). Yansong Qu, Dian Chen, Xinyang Li, Xiaofan Li, Shengchuan Zhang, Liujuan Cao, and Rongrong Ji

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  12. [12]

    3dsceneeditor: Controllable 3d scene editing with gaussian splatting.arXiv preprint arXiv:2412.01583(2024). Xianghui Yang, Huiwen Shi, Bowen Zhang, Fan Yang, Jiacheng Wang, Hongxu Zhao, Xinhai Liu, Xinzhou Wang, Qingxiang Lin, Jiaao Yu, Lifu Wang, Jing Xu, Zebin He, Zhuo Chen, Sicong Liu, Junta Wu, Yihang Lian, Shaoxiong Yang, Yuhong Liu, Yong SIGGRAPH Co...

    work page arXiv 2024

  13. [13]

    Hunyuan3d-1.0: A unified framework for text-to-3d and image-to-3d generation

    Hunyuan3D 1.0: A Unified Framework for Text-to-3D and Image-to-3D Generation. arXiv:2411.02293 Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang

    work page arXiv

  14. [14]

    IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models

    Ip-adapter: Text com- patible image prompt adapter for text-to-image diffusion models.arXiv preprint arXiv:2308.06721(2023). Junliang Ye, Shenghao Xie, Ruowen Zhao, Zhengyi Wang, Hongyu Yan, Wenqiang Zu, Lei Ma, and Jun Zhu

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  15. [15]

    Nano3d: A training-free approach for efficient 3d editing without masks.arXiv preprint arXiv:2510.15019, 2025

    NANO3D: A Training-Free Approach for Efficient 3D Editing Without Masks.arXiv preprint arXiv:2510.15019(2025). Xianfang Zeng, Xin Chen, Zhongqi Qi, Wen Liu, Zibo Zhao, Zhibin Wang, Bin Fu, Yong Liu, and Gang Yu

    work page arXiv 2025

  16. [16]

    IEEE Transactions on Pattern Analysis and Machine Intelligence(2025)

    Stylizedgs: Controllable stylization for 3d gaussian splatting. IEEE Transactions on Pattern Analysis and Machine Intelligence(2025). Longwen Zhang, Ziyu Wang, Qixuan Zhang, Qiwei Qiu, Anqi Pang, Haoran Jiang, Wei Yang, Lan Xu, and Jingyi Yu. 2024b. Clay: A controllable large-scale generative model for creating high-quality 3d assets.ACM Transactions on G...

    work page 2025

  17. [17]

    Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation

    Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation.arXiv preprint arXiv:2501.12202(2025). Yang Zheng, Hao Tan, Kai Zhang, Peng Wang, Leonidas Guibas, Gordon Wetzstein, and Wang Yifan

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  18. [18]

    Zijun Zhou, Yingying Deng, Xiangyu He, Weiming Dong, and Fan Tang

    SplatPainter: Interactive Authoring of 3D Gaussians from 2D Edits via Test-Time Training.arXiv preprint arXiv:2512.05354(2025). Zijun Zhou, Yingying Deng, Xiangyu He, Weiming Dong, and Fan Tang

    work page arXiv 2025

  19. [19]

    Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan

    Multi- turn Consistent Image Editing.arXiv preprint arXiv:2505.04320(2025). Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan

    work page arXiv 2025

  20. [20]

    ACM Transactions on Graphics (TOG)43, 4 (2024), 1–12

    Tip-editor: An accurate 3d editor following both text-prompts and image-prompts. ACM Transactions on Graphics (TOG)43, 4 (2024), 1–12. Jingyu Zhuang, Chen Wang, Liang Lin, Lingjie Liu, and Guanbin Li

    work page 2024

  21. [21]

    InSIGGRAPH Asia 2023 Conference Papers

    Dreameditor: Text-driven 3d scene editing with neural fields. InSIGGRAPH Asia 2023 Conference Papers. 1–10. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA. VecSet-Edit: Unleashing Pre-trained LRM for Mesh Editing from Single Image•13 A Details of VecSet-based LRM The illustration of overall VecSet Encoding and Diffusion process can...

    work page 2023

  22. [22]

    3 D.1 LRM Backbone Classifier-free Guidance Scale (𝑟= 10).This parameter controls the alignment of the generated mesh with the condition image. 3Sensitivity experiments were conducted on a randomly selected50%subset of the Edit3D-Bench to reduce computational overhead SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA. VecSet-Edit: Unl...

    work page 2026

  23. [23]

    SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA

    More qualitative demonstration. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA. 18•Teng-Fang Hsiao, Bo-Kai Ruan, Yu-Lun Liu, and Hong-Han Shuai Fig

    work page 2026

  24. [24]

    SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA

    More qualitative demonstration. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA

    work page 2026