pith. machine review for the scientific record. sign in

arxiv: 2604.26943 · v1 · submitted 2026-04-29 · 💻 cs.CV

Recognition: unknown

ProcFunc: Function-Oriented Abstractions for Procedural 3D Generation in Python

Alexander Raistrick, David Yan, Dylan Li, Erich Liang, Hongyu Wen, Jack Nugent, Jia Deng, Karhan Kayan, Lingjie Mei, Meenal Parakh, Yiming Zuo

Authors on Pith no claims yet

Pith reviewed 2026-05-07 08:14 UTC · model grok-4.3

classification 💻 cs.CV
keywords procedural 3D generationPython libraryBlendersynthetic data generationvision-language modelsindoor room modelingcompositional materialsprocedural abstractions
0
0 comments X

The pith

ProcFunc supplies Python functions that simplify creating and combining procedural 3D generation code for Blender.

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

The paper presents ProcFunc as a library of easy-to-use Python functions for Blender that streamline the creation, combination, analysis, and execution of procedural 3D generation code. This setup supports creating large-scale diverse training data through combinatorial compositions of semantic components. VLMs benefit by being able to edit and generate new procedural code with significantly reduced coding errors. The approach is illustrated by building a procedural indoor room generator that incorporates new compositional procedural materials and produces detailed, efficient, and diverse 3D synthetic data.

Core claim

ProcFunc provides a library of easy-to-use Python functions, which streamline creating, combining, analyzing, and executing procedural generation code. This makes it easy to create large-scale diverse training data, by combinatorial compositions of semantic components. VLMs can use ProcFunc to edit procedural material and geometry code and can create new procedural code with significantly fewer coding errors. Finally, as an example use case, we use ProcFunc to develop a new procedural generator of indoor rooms, which includes a collection of new compositional procedural materials. We demonstrate the detail, runtime efficiency, and diversity of this room generator, as well as its use for 3D合成

What carries the argument

The ProcFunc library of reusable Python functions that abstract Blender procedural capabilities for modular creation and execution of 3D generation scripts.

Load-bearing premise

The library's Python functions are easy enough for non-experts and VLMs to use effectively without quantitative evidence or comparisons to support the claims of fewer errors and high performance.

What would settle it

A test where VLMs generate procedural code for the same task using and not using the ProcFunc library, followed by measuring the rate of successful error-free executions and visual correctness of the outputs.

Figures

Figures reproduced from arXiv: 2604.26943 by Alexander Raistrick, David Yan, Dylan Li, Erich Liang, Hongyu Wen, Jack Nugent, Jia Deng, Karhan Kayan, Lingjie Mei, Meenal Parakh, Yiming Zuo.

Figure 1
Figure 1. Figure 1: ProcFunc Primitives API - ProcFunc provides a new API for low-level graphics operations which eliminates all non-essential or GUI-oriented complexity. Typ￾ical usage of the Blender and Infinigen API requires separate selection and action steps to manipulate Blender’s global background state. Blender and Infinigen require run￾time fields like “Base Color”, and “0, 6, 7” in the top row, which change dynamica… view at source ↗
Figure 2
Figure 2. Figure 2: Compositional Procedural Generation - ProcFunc enables reusable proce￾dural generators which can form countless semantically meaningful combinations. For materials, one can create bricks, tiles, paint and scratches combined with any material, forming tens of thousands of combinations. tate a GUI-based workflow: these steps are intuitive for artists and GUI-centric plugins, but can be significantly simplifi… view at source ↗
Figure 1
Figure 1. Figure 1: This leads to complex procedural code, where one function can fail silently view at source ↗
Figure 3
Figure 3. Figure 3: Example Indoor Room Generator - ProcFunc materials and scene arrange￾ment primitives can create useful scenes and datasets. Our tools can create high-detail room meshes in 1 to 2 minutes of CPU time. execution of the same random seed, or can cheaply execute all possible paths in less than a second, due to memoization and skipping execution of the internals of each mesh primitive. 3.3 Pre-implemented Proced… view at source ↗
Figure 4
Figure 4. Figure 4: Resource Usage: Our system provides a new and useful tradeoff of perfor￾mance and complexity. It matches the object count and triangle density of Infinigen￾Indoors, while being significantly more efficient on runtime, memory, and disk space. Our system also has the practical benefit of more predictable resource usage. We also compute quantitative measures of diversity in Tab. 3. We use the tracer (Sec 3.2)… view at source ↗
Figure 5
Figure 5. Figure 5: Complexity vs Efficiency: Proc￾Func enables efficient configurations for both high-detail scenes and for low-detail efficient scenes. We aim to evaluate the resource￾efficiency of the example procedural room generator (Sec 3.3). Efficiency is generally a tradeoff with scene de￾tail and complexity: higher triangle￾count or object-count scenes require both more memory and render time on the CPU. The best tra… view at source ↗
Figure 6
Figure 6. Figure 6: Geometry Detail - Ours produces more varied distribution of surface normal ground truth. ProcFunc’s provided material generators are designed to create highly detailed meshes by compositionally combining per-vertex dis￾placement shaders. Besides providing excel￾lent visual realism ( view at source ↗
read the original abstract

We introduce ProcFunc, a library for Blender-based procedural 3D generation in Python. ProcFunc provides a library of easy-to-use Python functions, which streamline creating, combining, analyzing, and executing procedural generation code. ProcFunc makes it easy to create large-scale diverse training data, by combinatorial compositions of semantic components. VLMs can use ProcFunc to edit procedural material and geometry code and can create new procedural code with significantly fewer coding errors. Finally, as an example use case, we use ProcFunc to develop a new procedural generator of indoor rooms, which includes a collection of new compositional procedural materials. We demonstrate the detail, runtime efficiency, and diversity of this room generator, as well as its use for 3D synthetic data generation. Please visit https://github.com/princeton-vl/procfunc for source code.

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 ProcFunc, a Python library for Blender-based procedural 3D generation. It provides a collection of easy-to-use functions intended to streamline the creation, combination, analysis, and execution of procedural code. The authors claim that VLMs can leverage ProcFunc to edit procedural material and geometry code and to generate new procedural code with significantly fewer coding errors. As an example application, the paper describes the development of a new procedural generator for indoor rooms that incorporates compositional procedural materials, and asserts that this generator exhibits high detail, runtime efficiency, and diversity, making it suitable for 3D synthetic data generation. Source code is made available via GitHub.

Significance. If the usability claims and performance assertions hold, ProcFunc could reduce the effort required to produce large-scale, combinatorially diverse 3D procedural content and synthetic datasets, which would benefit training of vision models. The open release of the implementation is a concrete strength that aids reproducibility and potential adoption. At present, however, the absence of any quantitative evaluation makes it difficult to determine whether these benefits are realized.

major comments (2)
  1. [Abstract] Abstract: the claim that VLMs 'can create new procedural code with significantly fewer coding errors' is unsupported. No controlled comparison, error taxonomy (syntax, runtime, semantic), or numerical error rates with versus without the library are provided.
  2. [Abstract] Abstract: the assertions of 'detail, runtime efficiency, and diversity' for the indoor-room generator lack any supporting metrics, tables, or statistical comparisons (e.g., render-time distributions, number of unique configurations, diversity measures such as Chamfer distance or perceptual hashes, or baselines against prior procedural room generators). Visual examples alone cannot substantiate these load-bearing claims.
minor comments (2)
  1. The manuscript would benefit from an explicit related-work section situating ProcFunc relative to existing Blender Python APIs, procedural modeling libraries, and VLM-assisted code-generation tools.
  2. A concise API overview or usage example with code snippets in the main text (beyond the GitHub link) would improve accessibility for readers who do not immediately consult the repository.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and for recognizing the potential utility of ProcFunc for procedural 3D generation and synthetic data creation. We agree that the abstract contains claims that are not quantitatively supported in the manuscript. We address each major comment below and will revise the abstract accordingly to ensure all statements are accurately grounded in the presented material.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that VLMs 'can create new procedural code with significantly fewer coding errors' is unsupported. No controlled comparison, error taxonomy (syntax, runtime, semantic), or numerical error rates with versus without the library are provided.

    Authors: We agree that the claim of VLMs creating new procedural code with significantly fewer coding errors is unsupported by any controlled comparison, error taxonomy, or quantitative rates in the manuscript. This phrasing was based on informal observations during our internal testing rather than systematic evaluation. We will revise the abstract to remove the phrase 'with significantly fewer coding errors.' The revised text will instead note that ProcFunc's high-level function abstractions simplify procedural code, which can facilitate VLM-assisted editing and generation of such code. This change preserves the intended point about the library's design while eliminating the unsubstantiated quantitative implication. revision: yes

  2. Referee: [Abstract] Abstract: the assertions of 'detail, runtime efficiency, and diversity' for the indoor-room generator lack any supporting metrics, tables, or statistical comparisons (e.g., render-time distributions, number of unique configurations, diversity measures such as Chamfer distance or perceptual hashes, or baselines against prior procedural room generators). Visual examples alone cannot substantiate these load-bearing claims.

    Authors: We concur that the abstract's assertions of 'detail, runtime efficiency, and diversity' for the indoor-room generator are not supported by quantitative metrics, tables, or comparisons against baselines. The current manuscript presents these qualities through visual examples and qualitative description of the compositional materials and scene generation process. We will revise the abstract to remove these specific load-bearing claims. The updated wording will focus on the generator's use of compositional procedural materials and its demonstration for 3D synthetic data generation via the provided examples, without asserting unsubstantiated levels of detail, efficiency, or diversity. If the revised manuscript length allows, we can add basic descriptive statistics (such as the number of unique room configurations generated) in the main text, but we will not introduce new quantitative evaluations or comparisons. revision: yes

Circularity Check

0 steps flagged

No circularity: descriptive library introduction without derivations or predictions

full rationale

The paper introduces ProcFunc as a Python library for Blender-based procedural 3D generation, describing its functions for creating/combining/analyzing/executing code and demonstrating an indoor-room generator via qualitative examples. No equations, fitted parameters, predictions, or derivation chains exist in the abstract or described content. Claims about VLM error reduction and generator properties (detail, efficiency, diversity) are presented as outcomes of using the library rather than derived quantities that reduce to inputs by construction. No self-citations, uniqueness theorems, or ansatzes are invoked. The work is self-contained as a tool paper; external benchmarks or quantitative metrics are absent but irrelevant to circularity analysis since no load-bearing derivation reduces to itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a software library and demonstration paper in computer graphics. It introduces no mathematical free parameters, no domain axioms beyond standard Python and Blender usage, and no new invented physical or theoretical entities. The contribution is the library code and its described usage patterns.

pith-pipeline@v0.9.0 · 5472 in / 1356 out tokens · 67617 ms · 2026-05-07T08:14:18.511506+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

38 extracted references · 20 canonical work pages

  1. [1]

    Murray, Benoit Steiner, Paul Tucker, Vijay Vasudevan, Pete Warden, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng

    Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghe- mawat, S., Irving, G., Isard, M., Kudlur, M., Levenberg, J., Monga, R., Moore, S., Murray, D.G., Steiner, B., Tucker, P.A., Vasudevan, V., Warden, P., Wicke, M., Yu, Y., Zhang, X.: Tensorflow: A system for large-scale machine learning. CoRR abs/1605.08695(2016),http://arxiv.or...

    work page arXiv 2016

  2. [2]

    com/jax-ml/jax3

    Bradbury, J., Frostig, R., Hawkins, P., Johnson, M.J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., VanderPlas, J., Wanderman-Milne, S., Zhang, Q.: JAX: composabletransformationsofPython+NumPyprograms(2018),http://github. com/jax-ml/jax3

    2018

  3. [3]

    Blender Foun- dation, Stichting Blender Foundation, Amsterdam (2018),http://www.blender

    Community, B.O.: Blender - a 3D modelling and rendering package. Blender Foun- dation, Stichting Blender Foundation, Amsterdam (2018),http://www.blender. org1

    2018

  4. [4]

    Deitke, M., VanderBilt, E., Herrasti, A., Weihs, L., Salvador, J., Ehsani, K., Han, W., Kolve, E., Farhadi, A., Kembhavi, A., Mottaghi, R.: ProcTHOR: Large-Scale EmbodiedAIUsingProceduralGeneration.In:NeurIPS(2022),outstandingPaper Award 4

    2022

  5. [5]

    Journal of Open Source Software (2024) 4

    Eppner, C., Murali, A., Garrett, C., O’Flaherty, R., Hermans, T., Yang, W., Fox, D.: scene_synthesizer: A python library for procedural scene generation in robot manipulation. Journal of Open Source Software (2024) 4

    2024

  6. [6]

    Ganeshan,A.,Huang,R.Y.,Xu,X.,Jones,R.K.,Ritchie,D.:Parsel:Parameterized shape editing with language (2024),https://arxiv.org/abs/2405.203195

    work page arXiv 2024

  7. [7]

    017864, 5, 10

    Gu, Y., Huang, I., Je, J., Yang, G., Guibas, L.: Blendergym: Benchmarking founda- tional model systems for graphics editing (2025),https://arxiv.org/abs/2504. 017864, 5, 10

    2025

  8. [8]

    Gumin, M., Han, D.H., Yoo, S.J., Ganeshan, A., Jones, R.K., Fu, K., Aguina- Kang, R., Morris, S., Ritchie, D.: Procedural scene programs for open-universe scene generation: Llm-free error correction via program search (2025),https:// arxiv.org/abs/2510.161475

    work page arXiv 2025

  9. [9]

    Huang, I., Bao, Y., Truong, K., Zhou, H., Schmid, C., Guibas, L., Fathi, A.: Fire- place: Geometric refinements of llm common sense reasoning for 3d object place- ment (2025),https://arxiv.org/abs/2503.049194

    work page arXiv 2025

  10. [10]

    arXiv preprint arXiv:2404.17672 (2024) 1, 4, 9

    Huang, I., Yang, G., Guibas, L.: Blenderalchemy: Editing 3d graphics with vision- language models. arXiv preprint arXiv:2404.17672 (2024) 1, 4, 9

    work page arXiv 2024

  11. [11]

    Jones, R.K., Guerrero, P., Mitra, N.J., Ritchie, D.: Shapelib: Designing a library of programmatic 3d shape abstractions with large language models (2025),https: //arxiv.org/abs/2502.088845

    work page arXiv 2025

  12. [12]

    The International Journal of Robotics Research , author =

    Karaman, S., Frazzoli, E.: Sampling-based algorithms for optimal motion plan- ning. The International Journal of Robotics Research30(7), 846–894 (2011). https : / / doi . org / 10 . 1177 / 0278364911406761,https : / / doi . org / 10 . 1177 / 0278364911406761, _eprint: https://doi.org/10.1177/0278364911406761 9

    work page doi:10.1177/0278364911406761 2011

  13. [13]

    The annual research report (1998),https://api.semanticscholar.org/CorpusID: 147446219 16 Raistrick et al

    LaValle, S.M.: Rapidly-exploring random trees: a new tool for path planning. The annual research report (1998),https://api.semanticscholar.org/CorpusID: 147446219 16 Raistrick et al

    1998

  14. [14]

    Li, B., Wu, R., Solar-Lezama, A., Zheng, C., Shi, L., Bickel, B., Matusik, W.: Vl- material: Procedural material generation with large vision-language models (2025), https://arxiv.org/abs/2501.186234

    work page arXiv 2025

  15. [15]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

    Lipson, L., Teed, Z., Deng, H., Ramanan, D.: Raft-stereo: Multilevel recurrent field transforms for stereo matching. In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). pp. 2501–2510 (2021) 14

    2021

  16. [16]

    Liu, X., Tang, C.K., Tai, Y.W.: Worldcraft: Photo-realistic 3d world creation and customization via llm agents (2025),https://arxiv.org/abs/2502.156014

    work page arXiv 2025

  17. [17]

    Lu, S., Chen, G., Dinh, N.A., Lang, I., Holtzman, A., Hanocka, R.: Ll3m: Large language 3d modelers (2025),https://arxiv.org/abs/2508.082281, 4

    work page arXiv 2025

  18. [18]

    ACM Trans

    Müller, P., Wonka, P., Haegler, S., Ulmer, A., Van Gool, L.: Procedural modeling of buildings. ACM Trans. Graph.25(3), 614–623 (Jul 2006).https://doi.org/ 10.1145/1141911.1141931,https://doi.org/10.1145/1141911.11419315

    work page doi:10.1145/1141911.1141931 2006

  19. [19]

    Nugent, J., Wu, S., Ma, Z., Han, B., Parakh, M., Joshi, A., Mei, L., Raistrick, A., Li, X., Deng, J.: Evaluating robustness of monocular depth estimation with procedural scene perturbations (2025),https://arxiv.org/abs/2507.009811

    work page arXiv 2025

  20. [20]

    In: 2012 IEEE International Conference on Robotics and Au- tomation

    Pan, J., Chitta, S., Manocha, D.: Fcl: A general purpose library for collision and proximity queries. In: 2012 IEEE International Conference on Robotics and Au- tomation. pp. 3859–3866 (2012).https://doi.org/10.1109/ICRA.2012.6225337 9

    work page doi:10.1109/icra.2012.6225337 2012

  21. [21]

    In: Wallach, H., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E., Garnett, R

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., Desmaison, A., Kopf, A., Yang, E., DeVito, Z., Raison, M., Tejani, A., Chilamkurthy, S., Steiner, B., Fang, L., Bai, J., Chintala, S.: Pytorch: An imperative style, high-performance deep learning library. In: Wallach, H., Larochelle, H....

    2019

  22. [22]

    ACM SIGGRAPH Computer Graphics (1985)

    Perlin, K.: An image synthesizer. ACM SIGGRAPH Computer Graphics (1985). https://doi.org/10.1145/325165.3252471

    work page doi:10.1145/325165.3252471 1985

  23. [23]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Raistrick, A., Lipson, L., Ma, Z., Mei, L., Wang, M., Zuo, Y., Kayan, K., Wen, H., Han, B., Wang, Y., Newell, A., Law, H., Goyal, A., Yang, K., Deng, J.: Infinite pho- torealistic worlds using procedural generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12630–12641 (2023) 1, 4, 5

    2023

  24. [24]

    In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR)

    Raistrick, A., Mei, L., Kayan, K., Yan, D., Zuo, Y., Han, B., Wen, H., Parakh, M., Alexandropoulos, S., Lipson, L., Ma, Z., Deng, J.: Infinigen indoors: Photorealistic indoor scenes using procedural generation. In: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR). pp. 21783–21794 (June 2024) 1, 4

    2024

  25. [25]

    arXiv preprint arXiv:2310.12945 (2023) 4

    Sun, C., Han, J., Deng, W., Wang, X., Qin, Z., Gould, S.: 3d-gpt: Procedural 3d modeling with large language models. arXiv preprint arXiv:2310.12945 (2023) 4

    work page arXiv 2023

  26. [26]

    Sun, F.Y., Liu, W., Gu, S., Lim, D., Bhat, G., Tombari, F., Li, M., Haber, N., Wu, J.: Layoutvlm: Differentiable optimization of 3d layout via vision-language models (2025),https://arxiv.org/abs/2412.021934

    work page arXiv 2025

  27. [27]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Sun, F.Y., Liu, W., Gu, S., Lim, D., Bhat, G., Tombari, F., Li, M., Haber, N., Wu, J.: Layoutvlm: Differentiable optimization of 3d layout via vision-language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 29469–29478 (June 2025) 4 ProcFunc 17

    2025

  28. [28]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Sun, J., Li, Y., Wei, J., Xu, L., Wang, N., Zhang, Y., Lu, C.: Arti-pg: A tool- box for procedurally synthesizing large-scale and diverse articulated objects with rich annotations. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 6396–6405 (2025) 4

    2025

  29. [29]

    Linux Journal2006(146), 10 (2006) 9

    Tomar, S.: Converting video formats with ffmpeg. Linux Journal2006(146), 10 (2006) 9

    2006

  30. [30]

    Wang, H., Xue, Q., Gao, W.: Infinibench: Infinite benchmarking for visual spatial reasoning with customizable scene complexity (2025),https://arxiv.org/abs/ 2511.182001

    work page arXiv 2025

  31. [31]

    Wang, W., Zhu, D., Wang, X., Hu, Y., Qiu, Y., Wang, C., Hu, Y., Kapoor, A., Scherer, S.: Tartanair: A dataset to push the limits of visual slam (2020) 9

    2020

  32. [32]

    arXiv (2025) 14

    Wen, B., Trepte, M., Aribido, J., Kautz, J., Gallo, O., Birchfield, S.: Foundation- stereo: Zero-shot stereo matching. arXiv (2025) 14

    2025

  33. [33]

    Yan, D., Raistrick, A., Deng, J.: What makes good synthetic training data for zero-shot stereo matching? (2026),https://arxiv.org/abs/2504.169301, 14

    work page arXiv 2026

  34. [34]

    Yang, Y., Jia, B., Zhang, S., Huang, S.: Sceneweaver: All-in-one 3d scene synthesis with an extensible and self-reflective agent (2025),https://arxiv.org/abs/2509. 204144

    2025

  35. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Yang, Y., Sun, F.Y., Weihs, L., VanderBilt, E., Herrasti, A., Han, W., Wu, J., Haber, N., Krishna, R., Liu, L., Callison-Burch, C., Yatskar, M., Kembhavi, A., Clark, C.: Holodeck: Language guided generation of 3d embodied ai environments. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 16227–16237 (June 2024) 4

    2024

  36. [36]

    Zhang, Y., Wang, Y., Zhang, Z., Tang, H.: Code2worlds: Empowering coding llms for 4d world generation (2026),https://arxiv.org/abs/2602.117574

    work page arXiv 2026

  37. [37]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Zhang, Y., Li, Z., Zhou, M., Wu, S., Wu, J.: The scene language: Representing scenes with programs, words, and embeddings. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 24625–24634 (2025) 5

    2025

  38. [38]

    Zhou, M., Wang, X., Wang, Y., Zhang, Z.: Roomcraft: Controllable and complete 3d indoor scene generation (2025),https://arxiv.org/abs/2506.222914

    work page arXiv 2025