pith. sign in

arxiv: 2606.10478 · v1 · pith:QM4ATL7Rnew · submitted 2026-06-09 · 💻 cs.CV

3D-CoS: A New 3D Reconstruction Paradigm Based on VLM Code Synthesis

Yuhao Wang , Puyi Wang , Linjie Li , Zhengyuan Yang , Kevin Qinghong Lin , Yu Cheng This is my paper

Pith reviewed 2026-06-27 14:03 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D reconstructioncode synthesisvision-language modelsBlender3D editingcontrollabilitylocalityVLM
0
0 comments X

The pith

3D objects reconstructed as executable Blender code enable more precise localized edits than point clouds or meshes.

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

The paper proposes representing 3D assets as executable Blender Python code generated by vision-language models instead of using implicit or explicit low-level forms such as NeRF, point clouds, or meshes. This code-based approach is evaluated across open and closed VLMs using workflows that include blueprint planning, retrieval over API docs, few-shot examples, and component-wise agent generation. The central evaluation compares code-based reconstruction to point-cloud baselines on text-driven editing tasks. Results indicate that the code format supplies stronger controllability and spatial locality, producing edits that match targets more closely while leaving untouched regions intact. The work positions code synthesis as a route toward programmatically editable 3D reconstruction.

Core claim

We introduce 3D-CoS, a reconstruction paradigm in which 3D assets are built as executable Blender code through VLM-driven synthesis. Systematic tests of representative VLMs and structured generation workflows demonstrate that code offers strong controllability and locality. In targeted editing evaluations this yields higher edit fidelity and better preservation of unedited regions than a point-cloud baseline.

What carries the argument

Executable Blender code as the 3D representation, which functions as both reconstruction output and an interpretable, directly editable programmatic medium.

If this is right

  • Text-driven edits can be applied to specific object parts with high fidelity.
  • Unedited regions remain unchanged more reliably than under point-cloud editing.
  • Workflows combining planning, RAG, and agent decomposition raise the success rate of code generation.
  • The same code representation supports both reconstruction and subsequent programmatic modification.

Where Pith is reading between the lines

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

  • Improved VLM code-generation ability would directly expand the range of scenes that can be reconstructed this way.
  • The format could be combined with existing scripting pipelines in animation or CAD tools.
  • Performance on very large or highly detailed scenes would test whether locality advantages scale.

Load-bearing premise

Vision-language models can produce functionally correct and complete Blender code that accurately captures the geometry and appearance of input objects.

What would settle it

A rendering test in which the generated Blender scripts fail to execute without errors or produce visual output that deviates substantially from the source images or descriptions.

Figures

Figures reproduced from arXiv: 2606.10478 by Kevin Qinghong Lin, Linjie Li, Puyi Wang, Yu Cheng, Yuhao Wang, Zhengyuan Yang.

Figure 1
Figure 1. Figure 1: An overview of our 3D code modeling paradigm. The top workflow summarizes our core process: code synthesis via VLMs, and its subsequent evaluation. Our work treats code as a unified representation for 3D assets. (Left) We demonstrate its capability in reconstruction, generating high-fidelity objects from a single image. (Right) We highlight its advantages in editing, where code-driven edits achieve superio… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our VLM-based 3D reconstruction pipeline. Solid arrows show the image￾to-code-to-3D workflow; dashed arrows show optional context from visual-code examples, blueprints, RAG-based API retrieval, and part-wise Agent generation. Appendix A.4 provides example artifacts. parameters as input. This is the fastest setting, but it often misses fine details or mis-specifies parameters for complex shapes.… view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Reconstruction variant & Code editing results. (Left) Examples of our text-conditional reconstruction variant, which modifies an object based on a source image and a textual instruction. For the reconstruction variant, Source denotes the original object rather than an edited ground truth. (Right) A direct comparison of our code editing method against the BlendedPC [35] baseline. Code￾driven editing shows s… view at source ↗
Figure 5
Figure 5. Figure 5: Rendered multi-view images of 3D object Chair 0891 in ModelNet10, with depth and [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Failure cases demonstration. The generated objects preserve category-level semantics but [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Additional examples of the text-conditional reconstruction variant. The model generates a modified 3D asset based on a source image from ModelNet10 and a corresponding text instruction. Here Source denotes the original object rather than an edited ground truth. trade-off between efficiency and performance. In practice, this suggests that single-call prompting may be preferable in resource-sensitive scenari… view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of articulated objects. We demonstrate the articulation effects generated by our Blender script across three categories. These include the sliding translation of cabinet drawers, the multi-axis rotation of monitor screens, and the hinge-based rotation of toilet lids, all simulating real-world usage scenarios [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: User study interface for image editing evaluation. Each item presents one reference render, two anonymous edited variants in randomized left/right order, the edit prompt, and rating controls for instruction following and preservation of unedited regions. 1. Let all the rings on the pillar sink with gravity and fit together. 2. The ring handle on the side of this cup is too big and does not match the cup bo… view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative results for code editing. Each example shows an input and edited object from BlendNet-E with a shortened text instruction; complete instructions are listed below. Input Edited [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Complementary code editing examples on BlendNet-E. These examples cover additional objects and edit intents beyond [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
read the original abstract

Most recent 3D reconstruction and editing systems operate on implicit and explicit representations such as NeRF, point clouds, or meshes. While these representations enable high-fidelity rendering, they are fundamentally low-level and hard to control programmatically. In contrast, we propose and systematically evaluate a new 3D reconstruction paradigm, 3D Code Synthesis (3D-CoS), where 3D assets are constructed as executable Blender code, a programmatic and interpretable medium. To assess how well current VLMs can use code to represent 3D objects, we evaluate representative open-source and closed-source VLMs in code-based reconstruction under a unified protocol. We further introduce a suite of structured code-synthesis workflows, including blueprint-based planning, Retrieval-Augmented Generation (RAG) over Blender API documentation, few-shot geometric demonstrations, and a component-level Agent workflow for part-wise code generation. To demonstrate the unique advantages of this representation, we further evaluate localized text-driven modifications and compare our code-based edits with a point-cloud-based 3D editing baseline. Our study shows that code as a 3D representation offers strong controllability and locality, yielding stronger edit fidelity and better preservation of unedited regions in our targeted editing evaluation. Our work also analyzes the potential of this paradigm, delineates the current capability frontier of VLMs for programmatic 3D modeling, and highlights code synthesis as a promising direction for editable 3D reconstruction.

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 / 1 minor

Summary. The manuscript proposes 3D-CoS, a paradigm in which 3D assets are represented as executable Blender code synthesized by VLMs rather than implicit or explicit geometric representations. It evaluates open- and closed-source VLMs under unified protocols using workflows such as blueprint planning, RAG over Blender API docs, few-shot demonstrations, and component-level agents; it further compares code-based localized text-driven edits against a point-cloud baseline and claims superior controllability, locality, edit fidelity, and preservation of unedited regions.

Significance. If the VLM-generated code proves reliably executable and geometrically faithful, the paradigm could enable more interpretable and programmatically editable 3D assets than current NeRF/mesh/point-cloud methods. The work also maps the current capability frontier of VLMs for programmatic 3D modeling. However, the absence of any reported quantitative results on code success rates, compilation errors, or geometric deviation prevents assessment of whether the claimed editing advantages are attributable to the representation itself.

major comments (2)
  1. [Abstract] Abstract: the claim that 'code as a 3D representation offers strong controllability and locality, yielding stronger edit fidelity and better preservation of unedited regions' is load-bearing for the central contribution, yet the abstract (and therefore the evaluation) provides no metrics, dataset details, success rates, or error analysis to support it.
  2. [Evaluation section (implied by abstract)] The targeted editing evaluation rests on the untested assumption that VLM-generated Blender code is functionally correct and complete; without reported compilation success rates, runtime error statistics, or deviation from ground-truth geometry, advantages over the point-cloud baseline cannot be attributed to the code representation.
minor comments (1)
  1. [Abstract] The abstract refers to 'our study shows' and 'our targeted editing evaluation' without naming the specific VLMs, datasets, or quantitative protocol used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback emphasizing the need for stronger quantitative grounding of our claims. We will revise the manuscript to incorporate additional metrics, success rates, and clarifications on the evaluation protocol while preserving the core contribution of the code-based paradigm.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'code as a 3D representation offers strong controllability and locality, yielding stronger edit fidelity and better preservation of unedited regions' is load-bearing for the central contribution, yet the abstract (and therefore the evaluation) provides no metrics, dataset details, success rates, or error analysis to support it.

    Authors: We agree the abstract should summarize supporting quantitative evidence. The evaluation section already contains human-rated edit fidelity scores, locality assessments, and side-by-side comparisons across 50+ editing examples on both open- and closed-source VLMs; we will condense these into the abstract (e.g., reporting average fidelity gains and unedited-region preservation rates) along with dataset size and workflow details. This change will make the load-bearing claim directly traceable to the reported results. revision: yes

  2. Referee: [Evaluation section (implied by abstract)] The targeted editing evaluation rests on the untested assumption that VLM-generated Blender code is functionally correct and complete; without reported compilation success rates, runtime error statistics, or deviation from ground-truth geometry, advantages over the point-cloud baseline cannot be attributed to the code representation.

    Authors: We acknowledge the importance of explicit success metrics. In revision we will add a new subsection reporting per-workflow compilation success rates, categorized runtime errors, and the fraction of generations that produced executable, renderable code. For the editing comparison we will restrict quantitative claims to the subset of successful code outputs and will state this filtering explicitly. Geometric deviation metrics are not directly available because many test cases start from textual descriptions rather than existing meshes; we will instead report functional equivalence (e.g., render consistency before/after edit) and clarify this scope limitation. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical evaluation of code-based 3D paradigm is self-contained

full rationale

The paper proposes 3D-CoS as a new paradigm using executable Blender code for 3D assets, introduces workflows (RAG, few-shot, agent), and reports empirical comparisons of editing fidelity against a point-cloud baseline. No equations, fitted parameters presented as predictions, or self-citation chains appear in the text. The controllability and locality claims are grounded in the described evaluation protocol rather than reducing to inputs by definition or prior self-work. This is a standard non-circular empirical proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only; no free parameters, invented entities, or non-standard axioms are described. The central claim rests on the domain assumption that VLMs can produce usable Blender code.

axioms (1)
  • domain assumption VLMs can be prompted to generate executable Blender code that accurately represents 3D geometry
    The evaluation and editing claims presuppose this capability exists at a useful level.

pith-pipeline@v0.9.1-grok · 5806 in / 1138 out tokens · 18090 ms · 2026-06-27T14:03:56.685937+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

81 extracted references · 16 canonical work pages · 10 internal anchors

  1. [1]

    Blendermcp - blender model context protocol integration, 2025

    Siddharth Ahuja and BlenderMCP Contributors. Blendermcp - blender model context protocol integration, 2025. URLhttps://github.com/ahujasid/blender-mcp

    2025

  2. [2]

    Introduction to model context protocol, 2024

    Anthropic PBC. Introduction to model context protocol, 2024. URL https://www. anthropic.com/news/model-context-protocol

    2024

  3. [3]

    Claude Sonnet 4, 2025

    Anthropic PBC. Claude Sonnet 4, 2025. URL https://www.anthropic.com/claude/ sonnet. Product page

    2025

  4. [4]

    Qwen2.5-VL Technical Report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. Qwen2.5-vl technical report. a...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Blender Python API reference, 2025

    Blender Foundation. Blender Python API reference, 2025. URL https://docs.blender. org/api/current/index.html. Online; accessed 2025-09-18

    2025

  6. [6]

    Blender: Open-source 3d creation suite

    Blender Online Community. Blender: Open-source 3d creation suite. https://www.blender. org, 2025. Version 4.4 as used in this work

    2025

  7. [7]

    ShapeNet: An Information-Rich 3D Model Repository

    Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shapenet: An information-rich 3d model repository, 2015. URL https://arxiv.org/abs/ 1512.03012

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  8. [8]

    Decor-gan: 3d shape detailization by conditional refinement

    Zhiqin Chen, Vladimir G Kim, Matthew Fisher, Noam Aigerman, Hao Zhang, and Siddhartha Chaudhuri. Decor-gan: 3d shape detailization by conditional refinement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15740–15749, 2021

    2021

  9. [9]

    Text-to-3d using gaussian splatting

    Zilong Chen, Feng Wang, Yikai Wang, and Huaping Liu. Text-to-3d using gaussian splatting. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 21401–21412, 2024

    2024

  10. [10]

    Meshcoder: Llm-powered structured mesh code generation from point clouds.arXiv preprint arXiv:2508.14879, 2025

    Bingquan Dai, Li Ray Luo, Qihong Tang, Jie Wang, Xinyu Lian, Hao Xu, Minghan Qin, Xudong Xu, Bo Dai, Haoqian Wang, et al. Meshcoder: Llm-powered structured mesh code generation from point clouds.arXiv preprint arXiv:2508.14879, 2025

    work page arXiv 2025

  11. [11]

    Blenderllm: Training large language models for computer-aided design with self-improvement, 2024

    Yuhao Du, Shunian Chen, Wenbo Zan, Peizhao Li, Mingxuan Wang, Dingjie Song, Bo Li, Yan Hu, and Benyou Wang. Blenderllm: Training large language models for computer-aided design with self-improvement, 2024. URLhttps://arxiv.org/abs/2412.14203

    work page arXiv 2024

  12. [12]

    Unreal engine c++ api reference, 2025

    Epic. Unreal engine c++ api reference, 2025. URL https://dev.epicgames.com/ documentation/en-us/unreal-engine/API

    2025

  13. [13]

    Unreal engine, 2025

    Epic. Unreal engine, 2025. URLhttps://www.unrealengine.com/

    2025

  14. [14]

    PAL: Program-aided Language Models

    Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. Pal: Program-aided language models, 2023. URL https://arxiv.org/ abs/2211.10435

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  15. [15]

    Gemini 3 Pro, 2025

    Google DeepMind. Gemini 3 Pro, 2025. URL https://blog.google/products/gemini/ gemini-3/. Official blog post (Nov 18, 2025)

    2025

  16. [16]

    Blendergym: Bench- marking foundational model systems for graphics editing

    Yunqi Gu, Ian Huang, Jihyeon Je, Guandao Yang, and Leonidas Guibas. Blendergym: Bench- marking foundational model systems for graphics editing. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 18574–18583, 2025

    2025

  17. [17]

    Scenecraft: An llm agent for synthesizing 3d scenes as blender code

    Ziniu Hu, Ahmet Iscen, Aashi Jain, Thomas Kipf, Yisong Yue, David A Ross, Cordelia Schmid, and Alireza Fathi. Scenecraft: An llm agent for synthesizing 3d scenes as blender code. In Forty-first International Conference on Machine Learning, 2024. 10

    2024

  18. [18]

    Blenderalchemy: Editing 3d graphics with vision-language models

    Ian Huang, Guandao Yang, and Leonidas Guibas. Blenderalchemy: Editing 3d graphics with vision-language models. InEuropean Conference on Computer Vision, pages 297–314. Springer, 2024

    2024

  19. [19]

    3d shape generation with grid-based implicit functions

    Moritz Ibing, Isaak Lim, and Leif Kobbelt. 3d shape generation with grid-based implicit functions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13559–13568, 2021

    2021

  20. [20]

    Kenny Jones, Theresa Barton, Xianghao Xu, Kai Wang, Ellen Jiang, Paul Guerrero, Niloy Mitra, and Daniel Ritchie

    R. Kenny Jones, Theresa Barton, Xianghao Xu, Kai Wang, Ellen Jiang, Paul Guerrero, Niloy Mitra, and Daniel Ritchie. Shapeassembly: Learning to generate programs for 3d shape structure synthesis.ACM Transactions on Graphics (TOG), Siggraph Asia 2020, 39(6):Article 234, 2020

    2020

  21. [21]

    Shap-E: Generating Conditional 3D Implicit Functions

    Heewoo Jun and Alex Nichol. Shap-e: Generating conditional 3d implicit functions.arXiv preprint arXiv:2305.02463, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  22. [22]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023

    2023

  23. [23]

    Nerf-vae: A geometry aware 3d scene generative model

    Adam R Kosiorek, Heiko Strathmann, Daniel Zoran, Pol Moreno, Rosalia Schneider, Sona Mokrá, and Danilo Jimenez Rezende. Nerf-vae: A geometry aware 3d scene generative model. InInternational conference on machine learning, pages 5742–5752. PMLR, 2021

    2021

  24. [24]

    Ln3diff: Scalable latent neural fields diffusion for speedy 3d generation

    Yushi Lan, Fangzhou Hong, Shuai Yang, Shangchen Zhou, Xuyi Meng, Bo Dai, Xingang Pan, and Chen Change Loy. Ln3diff: Scalable latent neural fields diffusion for speedy 3d generation. InEuropean Conference on Computer Vision, pages 112–130. Springer, 2024

    2024

  25. [25]

    LLaVA-OneVision: Easy Visual Task Transfer

    Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Yanwei Li, Ziwei Liu, and Chunyuan Li. Llava-onevision: Easy visual task transfer, 2024. URL https://arxiv.org/abs/2408.03326

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    Chain of code: Reasoning with a language model-augmented code emulator, 2024

    Chengshu Li, Jacky Liang, Andy Zeng, Xinyun Chen, Karol Hausman, Dorsa Sadigh, Sergey Levine, Li Fei-Fei, Fei Xia, and Brian Ichter. Chain of code: Reasoning with a language model-augmented code emulator, 2024. URLhttps://arxiv.org/abs/2312.04474

    work page arXiv 2024

  27. [27]

    Sp-gan: Sphere-guided 3d shape generation and manipulation.ACM Transactions on Graphics (TOG), 40(4):1–12, 2021

    Ruihui Li, Xianzhi Li, Ka-Hei Hui, and Chi-Wing Fu. Sp-gan: Sphere-guided 3d shape generation and manipulation.ACM Transactions on Graphics (TOG), 40(4):1–12, 2021

    2021

  28. [28]

    Code as Policies: Language Model Programs for Embodied Control

    Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. Code as policies: Language model programs for embodied control, 2023. URL https://arxiv.org/abs/2209.07753

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  29. [29]

    LL3M: Large language 3D modelers.arXiv:2508.08228, 2025

    Sining Lu, Guan Chen, Nam Anh Dinh, Itai Lang, Ari Holtzman, and Rana Hanocka. Ll3m: Large language 3d modelers.arXiv preprint arXiv:2508.08228, 2025

    work page arXiv 2025

  30. [30]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoor- thi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1):99–106, 2021

    2021

  31. [31]

    GPT-4o Technical Report.https://openai.com/index/hello-gpt-4o/, 2024

    OpenAI. GPT-4o Technical Report.https://openai.com/index/hello-gpt-4o/, 2024

    2024

  32. [32]

    Introducing OpenAI o3 and o4-mini, 2025

    OpenAI. Introducing OpenAI o3 and o4-mini, 2025. URL https://openai.com/index/ introducing-o3-and-o4-mini/. Model announcement (Apr 16, 2025)

    2025

  33. [33]

    DreamFusion: Text-to-3D using 2D Diffusion

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  34. [34]

    Qdrant: Vector database and vector search engine, 2025

    Qdrant Team. Qdrant: Vector database and vector search engine, 2025. URLhttps://github. com/qdrant/qdrant. GitHub repository, Version v1.15.4

    2025

  35. [35]

    Blended point cloud diffusion for localized text-guided shape editing, 2025

    Etai Sella, Noam Atia, Ron Mokady, and Hadar Averbuch-Elor. Blended point cloud diffusion for localized text-guided shape editing, 2025. URL https://arxiv.org/abs/2507.15399

    work page arXiv 2025

  36. [36]

    sphobjinv: A practical tool for manipulating sphinx objects.inv files, 2024

    Brian Skinn. sphobjinv: A practical tool for manipulating sphinx objects.inv files, 2024. URL https://github.com/bskinn/sphobjinv. 11

    2024

  37. [37]

    Sphinx documentation.https://www.sphinx-doc.org/, 2025

    The Sphinx Project. Sphinx documentation.https://www.sphinx-doc.org/, 2025

    2025

  38. [38]

    Freeman, Joshua B

    Yonglong Tian, Andrew Luo, Xingyuan Sun, Kevin Ellis, William T. Freeman, Joshua B. Tenenbaum, and Jiajun Wu. Learning to infer and execute 3d shape programs. InInternational Conference on Learning Representations, 2019

    2019

  39. [39]

    Unity real-time development platform, 2025

    Unity. Unity real-time development platform, 2025. URLhttps://unity.com/

    2025

  40. [40]

    Unity scripting api, 2025

    Unity. Unity scripting api, 2025. URL https://docs.unity3d.com/6000.2/ Documentation/ScriptReference/index.html

    2025

  41. [41]

    Clip-nerf: Text- and-image driven manipulation of neural radiance fields

    Can Wang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Clip-nerf: Text- and-image driven manipulation of neural radiance fields. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3835–3844, 2022

    2022

  42. [42]

    Rodin: A generative model for sculpting 3d digital avatars using diffusion

    Tengfei Wang, Bo Zhang, Ting Zhang, Shuyang Gu, Jianmin Bao, Tadas Baltrusaitis, Jingjing Shen, Dong Chen, Fang Wen, Qifeng Chen, et al. Rodin: A generative model for sculpting 3d digital avatars using diffusion. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4563–4573, 2023

    2023

  43. [43]

    InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency

    Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Shenglong Ye, Jie Shao, et al. Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency.arXiv preprint arXiv:2508.18265, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  44. [44]

    Karl D. D. Willis, Yewen Pu, Jieliang Luo, Hang Chu, Tao Du, Joseph G. Lambourne, Armando Solar-Lezama, and Wojciech Matusik. Fusion 360 gallery: A dataset and environment for programmatic cad construction from human design sequences.ACM Transactions on Graphics (TOG), 40(4), 2021

    2021

  45. [45]

    Unique3d: High-quality and efficient 3d mesh generation from a single image, 2024

    Kailu Wu, Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan, and Kaisheng Ma. Unique3d: High-quality and efficient 3d mesh generation from a single image, 2024

    2024

  46. [46]

    Deepcad: A deep generative network for computer- aided design models

    Rundi Wu, Chang Xiao, and Changxi Zheng. Deepcad: A deep generative network for computer- aided design models. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 6772–6782, October 2021

    2021

  47. [47]

    Textsplat: Text-guided semantic fusion for generalizable gaussian splatting

    Zhicong Wu, Hongbin Xu, Gang Xu, Ping Nie, Zhixin Yan, Jinkai Zheng, Liangqiong Qu, Ming Li, and Liqiang Nie. Textsplat: Text-guided semantic fusion for generalizable gaussian splatting. arXiv preprint arXiv:2504.09588, 2025

    work page arXiv 2025

  48. [48]

    3D ShapeNets: A Deep Representation for Volumetric Shapes

    Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Linguang Zhang, Xiaoou Tang, and Jianxiong Xiao. 3d shapenets: A deep representation for volumetric shapes, 2015. URL https://arxiv.org/abs/1406.5670

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  49. [49]

    InstantMesh: Efficient 3D Mesh Generation from a Single Image with Sparse-view Large Reconstruction Models

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. In- stantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruc- tion models.arXiv preprint arXiv:2404.07191, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  50. [50]

    Cad-mllm: Unifying multimodality-conditioned cad generation with mllm, 2024

    Jingwei Xu, Chenyu Wang, Zibo Zhao, Wen Liu, Yi Ma, and Shenghua Gao. Cad-mllm: Unifying multimodality-conditioned cad generation with mllm, 2024

    2024

  51. [51]

    Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models

    Taoran Yi, Jiemin Fang, Junjie Wang, Guanjun Wu, Lingxi Xie, Xiaopeng Zhang, Wenyu Liu, Qi Tian, and Xinggang Wang. Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6796–6807, 2024

    2024

  52. [52]

    Lion: Latent point diffusion models for 3d shape generation

    xiaohui zeng, Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, and Karsten Kreis. Lion: Latent point diffusion models for 3d shape generation. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neu- ral Information Processing Systems, volume 35, pages 10021–10039. Curran Associates, Inc., 2022. UR...

    2022

  53. [53]

    Sin." stands for “single-call

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models.ACM Transactions On Graphics (TOG), 42(4):1–16, 2023. 13 A Appendix A.1 Reconstruction Pipeline Details ModelNet10 Easy/Hard Split.We partition ModelNet10 intoeasyandhardsubsets per category: objects wit...

    2023

  54. [54]

    Let all the rings on the pillar sink with gravity and fit together

  55. [55]

    Make it smaller

    The ring handle on the side of this cup is too big and does not match the cup body. Make it smaller

  56. [56]

    Make it thinner and longer and reduce the number to 1 and insert it in the middle of the top of the cake

    The candle on this cake is too thick and short. Make it thinner and longer and reduce the number to 1 and insert it in the middle of the top of the cake

  57. [57]

    Change the frustum-shaped lampshade of the upper part of the table lamp into a cylindrical shape

  58. [58]

    Position the top layer of the burger off-center so people can see the insides

  59. [59]

    Turn it to the closed position

    This oil-paper umbrella is open, with a cone on top. Turn it to the closed position. In Figure 4, the instructions we use are:

  60. [60]

    Make the bathtub more square and add a flat base for stability. 19 Input Edited All the rings sink with gravity… Make the ring handle smaller… Make the candle one and taller… Make lampshade into cylindrical… Close the oil- paper umbrella… Position the top layer off-center… Figure 10:Qualitative results for code editing.Each example shows an input and edit...

  61. [61]

    Add a second drawer below the existing one

  62. [62]

    Change the base legs to a single centered pedestal

  63. [63]

    Replace the cylindrical lampshade above this desk lamp with a triangular cone

  64. [64]

    The column mistakenly passes through the lampshade and protrudes a little from the top

    Make this table lamp taller. The column mistakenly passes through the lampshade and protrudes a little from the top. Remove this small part

  65. [65]

    In Figure 7, the instructions we use are: • Upper part:

    Lengthen the four cylindrical legs of this table and connect the legs at opposite corners at the bottom with X-shaped wooden strips to make its structure more stable. In Figure 7, the instructions we use are: • Upper part:

  66. [66]

    Add a lower shelf between the two legs

  67. [67]

    Convert the corner bath to an oval shape

  68. [68]

    Convert one of the crib’s sides into a removable panel

  69. [69]

    Cut a large opening in the middle of the backrest

  70. [70]

    • Lower part:

    Extend the basin to double its current length. • Lower part:

  71. [71]

    Add a central open shelf in the knee space area for additional storage

  72. [72]

    Add a headboard to the bed

  73. [73]

    Add a fifth drawer at the bottom

  74. [74]

    Add a second, smaller screen on top to create a dual-monitor setup

  75. [75]

    In Figure 11, the instructions we use are:

    Add a lower central support beam between the sofa legs. In Figure 11, the instructions we use are:

  76. [76]

    This sofa has armrests on only one side and the modification makes it have armrests on both sides. 20

  77. [77]

    The keychain circle on this cup is too big; make it smaller

  78. [78]

    Make it hollow

    The cylindrical portion of this cup was incorrectly generated as a solid shape. Make it hollow

  79. [79]

    Add a handguard in the middle of this sofa to give it two separate seats

  80. [80]

    Separate the spherical part of this bulb from the base

Showing first 80 references.