pith. sign in

arxiv: 2606.29301 · v1 · pith:CPWOC3HHnew · submitted 2026-06-28 · 💻 cs.CV

Pointer-CAD v2: Plan-Then-Construct CAD Generation with Dimension-Aware Parametric Precision

Dacheng Qi , Chenyu Wang , Jingwei Xu , Yi Ma , Shenghua Gao This is my paper

Pith reviewed 2026-06-30 07:53 UTC · model grok-4.3

classification 💻 cs.CV
keywords CAD generationparametric precisionpointer mechanismdesign plancontinuous parametersgeometric accuracyLLM sequence prediction
0
0 comments X

The pith

Pointer-CAD v2 generates dimensionally precise CAD by first planning explicit metrics then referencing them via pointers during construction.

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

The paper establishes that CAD generation can achieve industrial-level parametric accuracy by decoupling the reasoning of exact numerical parameters from the geometric sequence construction. It first creates a structured design plan that lists metric scale values explicitly, stores them in a dictionary, and then uses a pointer mechanism to reference those values directly while generating the construction sequence. This removes the need to quantize numbers for LLM prediction, which previously introduced small errors that violate tolerance requirements even when shapes look similar. A new dataset with plan-level annotations and three hierarchical metrics for vertex, edge, and face accuracy enable direct measurement of this fidelity.

Core claim

Pointer-CAD v2 first produces a structured design plan containing explicit metric scale parameters, organizes those parameters into a dictionary, and then references them directly via a pointer mechanism while generating the command sequence. This allows continuous value prediction instead of quantized values, eliminates discretization errors, and produces dimensionally consistent CAD outputs that outperform prior baselines on geometric accuracy measures.

What carries the argument

The Plan-Then-Construct paradigm with a pointer mechanism that references explicit metric parameters from a dictionary during autoregressive sequence generation.

If this is right

  • CAD generation becomes suitable for precision-critical engineering tasks that require exact metric compliance.
  • Geometric accuracy improves measurably at vertex, edge, and face levels without relying on post-hoc corrections.
  • Quantization steps are eliminated, allowing direct output of continuous parameter values in sequence models.
  • Evaluation can shift from visual similarity to parametric fidelity using the new hierarchical metrics.

Where Pith is reading between the lines

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

  • The same plan-then-reference structure could reduce precision loss in other autoregressive generation domains such as 3D assembly or manufacturing code.
  • The hierarchical geometry metrics may generalize to other parametric modeling tasks where small deviations matter.
  • Robust plan generation would allow scaling to assemblies with many interdependent dimensions.

Load-bearing premise

An LLM can reliably produce a correct structured design plan with accurate explicit metric parameters that the pointer mechanism will reference without introducing mismatches or inconsistencies.

What would settle it

A test set where CAD outputs generated from the plan show vertex or edge dimensions that deviate beyond tolerance from the explicit metric values listed in the plan.

Figures

Figures reproduced from arXiv: 2606.29301 by Chenyu Wang, Dacheng Qi, Jingwei Xu, Shenghua Gao, Yi Ma.

Figure 1
Figure 1. Figure 1: Strength of Our Proposed Method and Metric. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Pointer-CAD v2 Pipeline. Pointer-CAD v2 generates CAD models in a step￾wise manner under a Plan-Then-Construct framework. Each step includes a plan stage and a construction stage. The plan stage produces a structured, dimension-aware design plan conditioned on the full prompt and the existing B-rep. The construction stage retrieves the planned parameters to form a command sequence, which is constructed to … view at source ↗
Figure 3
Figure 3. Figure 3: Example of a dimension-aware design plan. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Plan parameter encoding pipeline. All parameters are first extracted from the generated plan and organized into separate length and angle dictionaries. Each dictionary is processed by a parameter encoder that captures both frequency and con￾textual information. The encoder maps every parameter to a dedicated embedding, forming a set of parameter embeddings for subsequent modeling. referable parameters are … view at source ↗
Figure 5
Figure 5. Figure 5: Plan Construction Pipeline. Key parameters are extracted from raw JSON files and combined with expert annotations to form structured prompts. These prompts are fed into Qwen3 to generate detailed design plans. The generated plans are verified for parameter completeness. If required parameters are missing, Qwen3 is re-prompted up to two times, and any remaining failures are discarded. methods with different… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison with specialized text-to-CAD methods. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of generation and failure cases. [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Example of CAD model editing via plan modification. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation study on metric tolerance. We assess the average accuracy across three metric levels for different values of the tolerance factor k. Since performance remains stable when k is around 10−3 , we set k = 10−3 in all experiments. dataset construction, the plans in the Pointer-CAD v2 dataset and the parame￾ter books in the Pointer-CAD dataset both involve multiple format conversions of the parameters. … view at source ↗
Figure 10
Figure 10. Figure 10: GUI platform for manual inspection. This interface is used for manual inspection of the generated plans. Annotators can review expert annotations, required parameters with their values, and the generated plans. They can also directly revise the plans to ensure accuracy and consistency. these issues according to the syntax rules of the modeling operations to ensure that parameter IDs are consecutive and pa… view at source ↗
Figure 11
Figure 11. Figure 11: Dataset statistics. We analyze the distribution of modeling operations and the number of modeling steps per CAD model across multiple datasets, including Deep￾CAD [46], OmniCAD [49], Recap-OmniCAD+ [36], and OmniCAD-Plan+. 22.08% relative to Recap-OmniCAD+, whereas chamfer and fillet operations decrease by only 3.7%. We attribute this difference to the higher complexity of sketch-extrude pairs. Such opera… view at source ↗
Figure 12
Figure 12. Figure 12: Distribution of modeling steps. All datasets show a clear long-tail distri￾bution, with many models requiring only one modeling step. the necessary information for Qwen3 to understand the CAD model. The place￾holder INSERT_PROMPT_HERE is replaced with the expert-level model description from the source dataset. The placeholder INSERT_JSON_HERE is replaced with the required parameters and their correspondin… view at source ↗
Figure 13
Figure 13. Figure 13: System prompt for sketch-extrude operation pairs. [PITH_FULL_IMAGE:figures/full_fig_p029_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: System prompt for chamfer operation. This prompt provides a detailed description of the chamfer operation and specifies the expected structure of the gener￾ated design plan [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: System prompt for fillet operation. This prompt provides a detailed description of the fillet operation and specifies the expected structure of the generated design plan [PITH_FULL_IMAGE:figures/full_fig_p031_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: User prompt. This prompt specifies the parameters that should appear in the generated design plan. JSON Parameters for Sketch-Extrude Operation Pairs { "sketch plane": { "origin": {"x": 0.0, "y": 0.0, "z": 0.0}, "rotation": { "x to u": 0.0, "y to u": 270.0, "z to u": "ignore" } }, "curves": { "arc_0": { "endpoint1": {"x": 0, "y": 0}, "endpoint2": {"x": 1, "y": 1}, "sweep angle": 180 }, "line_1": { "endpoi… view at source ↗
Figure 17
Figure 17. Figure 17: JSON parameters for sketch-extrude operation pairs. [PITH_FULL_IMAGE:figures/full_fig_p032_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: JSON parameters for chamfer and fillet operations. [PITH_FULL_IMAGE:figures/full_fig_p033_18.png] view at source ↗
read the original abstract

Computer-aided design (CAD) plays a fundamental role in modern manufacturing by providing the high precision required for industrial production. Recent large language model based approaches formulate CAD generation as a sequence prediction problem and have achieved promising results. However, existing methods and evaluation protocols primarily emphasize visual similarity, while overlooking precise geometric parameters and correct metric scale. Small numerical deviations that are negligible at the shape-level may still violate industrial tolerance requirements, a problem further compounded by current autoregressive paradigms that utilize command sequence representations, aggressively quantize numerical parameters to ease LLM prediction. In this work, we present Pointer-CAD v2. Compared with v1 (arXiv:2603.04337), this version directly predicts continuous values, bypassing the need for quantized numerical parameters and thereby eliminating quantization errors. Specifically, we propose a unified framework that decouples parameter reasoning from geometric construction through a Plan-Then-Construct paradigm. Our method first produces a structured design plan with explicit metric scale parameters. These parameters are organized into a dictionary and directly referenced during sequence generation via a pointer mechanism, eliminating discretization errors and ensuring dimensionally consistent execution. In addition, we construct a new large-scale dataset with plan-level annotation and introduce three hierarchical geometry accuracy metrics to evaluate parametric fidelity at the vertex, edge, and face levels. Extensive experiments demonstrate that Pointer-CAD v2 consistently outperforms existing baselines and achieves substantial improvements in geometric accuracy, enabling reliable CAD generation for precision-critical engineering applications.

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 / 0 minor

Summary. Pointer-CAD v2 presents a Plan-Then-Construct framework for LLM-based CAD generation. It first produces a structured design plan containing explicit metric-scale parameters organized as a dictionary; these are then referenced by a pointer mechanism during autoregressive sequence construction to directly predict continuous values and eliminate quantization error. The work introduces a new large-scale dataset with plan-level annotations and three hierarchical geometry-accuracy metrics (vertex, edge, face) intended to evaluate parametric fidelity beyond visual similarity. Experiments are stated to show consistent outperformance over baselines for precision-critical applications.

Significance. If the plan-generation accuracy and pointer-consistency claims are substantiated, the decoupling of parameter reasoning from geometric construction and the avoidance of discretization could meaningfully advance LLM-driven CAD for engineering tolerances. The hierarchical metrics and annotated dataset would also provide useful evaluation tools for the community. The significance hinges on whether the reported geometric gains are attributable to the pointer mechanism rather than post-hoc metric design or unverified upstream plan quality.

major comments (2)
  1. [Abstract] Abstract: the central claim that the pointer mechanism 'eliminates quantization errors and ensures dimensionally consistent execution' rests on the unverified load-bearing assumption that the upstream structured design plan is generated correctly; no plan-level metrics (parameter correctness rate, reference error rate) are reported, so downstream geometric accuracy improvements cannot be isolated or confirmed.
  2. [Abstract] Abstract / Experiments: the statements 'consistently outperforms existing baselines' and 'achieves substantial improvements in geometric accuracy' are presented without any quantitative results, error bars, dataset statistics, or table references, preventing assessment of whether the hierarchical metrics actually demonstrate the claimed parametric gains under identical conditions for baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract and the need for stronger substantiation of our claims. We address each major comment below and commit to revisions that improve verifiability without altering the core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the pointer mechanism 'eliminates quantization errors and ensures dimensionally consistent execution' rests on the unverified load-bearing assumption that the upstream structured design plan is generated correctly; no plan-level metrics (parameter correctness rate, reference error rate) are reported, so downstream geometric accuracy improvements cannot be isolated or confirmed.

    Authors: We agree that plan-level metrics are necessary to isolate the pointer mechanism's contribution. The revised manuscript will include explicit plan-level metrics (parameter correctness rate and reference error rate) on the annotated dataset. These additions will allow direct verification of upstream plan quality and confirm that geometric gains arise from exact continuous parameter referencing rather than plan inaccuracies. revision: yes

  2. Referee: [Abstract] Abstract / Experiments: the statements 'consistently outperforms existing baselines' and 'achieves substantial improvements in geometric accuracy' are presented without any quantitative results, error bars, dataset statistics, or table references, preventing assessment of whether the hierarchical metrics actually demonstrate the claimed parametric gains under identical conditions for baselines.

    Authors: The abstract is a high-level summary; full quantitative results with tables, error bars, dataset statistics, and baseline comparisons under the new hierarchical metrics appear in the Experiments section. We will revise the abstract to incorporate key quantitative improvements and explicit table references, enabling readers to assess the parametric gains directly from the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents a Plan-Then-Construct framework with a pointer mechanism for referencing explicit parameters from a generated design plan, directly predicting continuous values instead of quantized ones, and introduces new hierarchical geometry accuracy metrics plus a dataset with plan-level annotations. No equations, derivations, or central claims reduce by construction to fitted inputs or self-citations; the v1 reference (arXiv:2603.04337) is used only for comparison. The new metrics evaluate parametric fidelity at vertex/edge/face levels but are not shown to be tuned in a self-definitional way that forces the reported gains. The method's claims rest on empirical outperformance under the proposed evaluation protocol rather than any self-referential loop or imported uniqueness theorem.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract alone supplies insufficient detail to enumerate concrete free parameters, axioms, or invented entities; the pointer mechanism and plan dictionary are methodological choices rather than new postulated physical entities or unproven mathematical axioms.

pith-pipeline@v0.9.1-grok · 5803 in / 1168 out tokens · 28907 ms · 2026-06-30T07:53:56.588555+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

64 extracted references · 18 canonical work pages · 5 internal anchors

  1. [1]

    com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf

    Anthropic: Claude opus 4.5 system card (2025),https://www-cdn.anthropic. com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf

    2025

  2. [2]

    https://doi.org/10.5281/zenodo

    CadQuery Contributors: CadQuery (2025),https://doi.org/10.5281/zenodo. 14590990

    work page doi:10.5281/zenodo 2025

  3. [3]

    In: CVPR

    Chen, C., Wei, J., Chen, T., Zhang, C., Yang, X., Zhang, S., Yang, B., Foo, C.S., Lin, G., Huang, Q., et al.: CADCrafter: Generating computer-aided design models from unconstrained images. In: CVPR. pp. 11073–11082 (2025)

    2025

  4. [4]

    Evaluating Large Language Models Trained on Code

    Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H.P.D.O., Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G., et al.: Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  5. [5]

    IEEE Transactions on Industrial Informatics (2025)

    Chen, T., Yu, C., Hu, Y., Li, J., Xu, T., Cao, R., Zhu, L., Zang, Y., Zhang, Y., Li, Z., et al.: Img2CAD: Conditioned 3-d cad model generation from single image with structured visual geometry. IEEE Transactions on Industrial Informatics (2025)

    2025

  6. [6]

    In: ICML

    Dauphin, Y.N., Fan, A., Auli, M., Grangier, D.: Language modeling with gated convolutional networks. In: ICML. pp. 933–941. PMLR (2017)

    2017

  7. [7]

    HistCAD: A Constraint-Aware Parametric History-Based CAD Representation, Dataset, and Benchmark with Industrial Complexity

    Dong, X., Li, C., Han, C., Zheng, P., Jing, J., Song, Y., Yang, Z.: HistCAD: Geometrically constrained parametric history-based cad dataset. arXiv preprint arXiv:2602.19171 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    In: ICML

    Du, Y., Durkan, C., Strudel, R., Tenenbaum, J.B., Dieleman, S., Fergus, R., Sohl- Dickstein, J., Doucet, A., Grathwohl, W.S.: Reduce, reuse, recycle: Compositional generation with energy-based diffusion models and mcmc. In: ICML. pp. 8489–

  9. [9]

    In: ECCV

    Dupont, E., Cherenkova, K., Mallis, D., Gusev, G., Kacem, A., Aouada, D.: Tran- sCAD: A hierarchical transformer for cad sequence inference from point clouds. In: ECCV. pp. 19–36. Springer (2024)

    2024

  10. [10]

    Cadevolve: Creating realistic cad via program evolution.arXiv preprint arXiv:2602.16317, 2026

    Elistratov,M.,Barannikov,M.,Ivanov,G.,Khrulkov,V.,Konushin,A.,Kuznetsov, A., Zhemchuzhnikov, D.: CADEvolve: Creating realistic cad via program evolution. arXiv preprint arXiv:2602.16317 (2026)

    work page arXiv 2026

  11. [11]

    In: ICML (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: ICML (2024)

    2024

  12. [12]

    Integrated Computer-Aided Engineering32(1), 75–96 (2025)

    Fan, R., He, F., Liu, Y., Song, Y., Fan, L., Yan, X.: A parametric and feature- based cad dataset to support human-computer interaction for advanced 3d shape learning. Integrated Computer-Aided Engineering32(1), 75–96 (2025)

    2025

  13. [13]

    FreeCAD Community: FreeCAD (2024),https://www.freecad.org/

    2024

  14. [14]

    Google DeepMind: Gemini 3 pro (2025),https://deepmind.google/models/ gemini/pro/

    2025

  15. [15]

    TMLR (2026)

    Govindarajan, P., Baldelli, D., Pathak, J., Fournier, Q., Chandar, S.: CADmium: Fine-tuning code language models for text- driven sequential cad design. TMLR (2026)

    2026

  16. [16]

    CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward

    Guan, Y., Wang, X., Xing, X., Zhang, J., Xu, D., Yu, Q.: CAD-Coder: Text- to-cad generation with chain-of-thought and geometric reward. arXiv preprint arXiv:2505.19713 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    In: ICLR (2026),https://openreview

    Guo,Q.,Dai,G.,Liu,Z.,Huang,S.,Hu,Y.,Zhang,H.,Chen,T.:PlanthenAct:Bi- level CAD command sequence generation. In: ICLR (2026),https://openreview. net/forum?id=LhE03r8X8z

    2026

  18. [18]

    In: Computer Graphics Forum

    Jones, B.T., Zhang, Z., Hähnlein, F., Matusik, W., Ahmad, M., Kim, V., Schulz, A.: A solver-aided hierarchical language for llm-driven cad design. In: Computer Graphics Forum. vol. 44, p. e70250. Wiley Online Library (2025) Pointer-CAD v2 17

    2025

  19. [19]

    NeurIPS35, 26565–26577 (2022)

    Karras,T.,Aittala,M.,Aila,T.,Laine,S.:Elucidatingthedesignspaceofdiffusion- based generative models. NeurIPS35, 26565–26577 (2022)

    2022

  20. [20]

    NeurIPS37, 7552–7579 (2024)

    Khan,M.S.,Sinha,S.,Sheikh,T.U.,Stricker,D.,Ali,S.A.,Afzal,M.Z.:Text2CAD: Generating sequential cad designs from beginner-to-expert level text prompts. NeurIPS37, 7552–7579 (2024)

    2024

  21. [21]

    In: CVPR

    Khan, M.S., Dupont, E., Ali, S.A., Cherenkova, K., Kacem, A., Aouada, D.: CAD- SIGNet: Cad language inference from point clouds using layer-wise sketch instance guided attention. In: CVPR. pp. 4713–4722 (2024)

    2024

  22. [22]

    cadrille: Multi-modal cad reconstruction with online reinforcement learning

    Kolodiazhnyi, M., Tarasov, D., Zhemchuzhnikov, D., Nikulin, A., Zisman, I., Vorontsova, A., Konushin, A., Kurenkov, V., Rukhovich, D.: cadrille: Multi- modal cad reconstruction with online reinforcement learning. arXiv preprint arXiv:2505.22914 (2025)

    work page arXiv 2025

  23. [23]

    arXiv preprint arXiv:2512.11199 (2025)

    Le, T., Nguyen, K., Huang, B., Ta, T.D., Nguyen, A.: CADKnitter: Compositional cad generation from text and geometry guidance. arXiv preprint arXiv:2512.11199 (2025)

    work page arXiv 2025

  24. [24]

    In: CVPR

    Li, J., Ma, W., Li, X., Lou, Y., Zhou, G., Zhou, X.: CAD-Llama: leveraging large language models for computer-aided design parametric 3d model generation. In: CVPR. pp. 18563–18573 (2025)

    2025

  25. [25]

    arXiv preprint arXiv:2601.09428 (2026)

    Li, S., Lambourne, J.G., Zhang, L., Jayaraman, P.K., Willis, K., et al.: Draw it like Euclid: Teaching transformer models to generate cad profiles using ruler and compass construction steps. arXiv preprint arXiv:2601.09428 (2026)

    work page arXiv 2026

  26. [26]

    Seek-cad: A self-refined generative modeling for 3d parametric cad using local inference via deepseek.arXiv preprint arXiv:2505.17702, 2025

    Li, X., Li, J., Song, Y., Lou, Y., Zhou, X.: Seek-CAD: A self-refined generative modeling for 3d parametric cad using local inference via deepseek. arXiv preprint arXiv:2505.17702 (2025)

    work page arXiv 2025

  27. [27]

    In: CVPR

    Li, Y., Lin, C., Liu, Y., Long, X., Zhang, C., Wang, N., Li, X., Wang, W., Guo, X.: CADDreamer: Cad object generation from single-view images. In: CVPR. pp. 21448–21457 (2025)

    2025

  28. [28]

    Journal of Computational Design and Engineering12(10), 176–184 (2025)

    Liang, J., He, F., Fan, R., Chu, Y., Yan, X.: CADCL: Reconstruct parametric cad models from b-rep via contrastive learning. Journal of Computational Design and Engineering12(10), 176–184 (2025)

    2025

  29. [29]

    In: ACM MM

    Lin, D., Yuan, M., Wang, Z., Wu, T., Liu, Y.: FreeCAD: A multimodal framework for 3d cad model generation from free-form prompts. In: ACM MM. pp. 1948–1956 (2025)

    1948

  30. [30]

    In: CVPR

    Ma, W., Chen, S., Lou, Y., Li, X., Zhou, X.: Draw step by step: Reconstructing cad construction sequences from point clouds via multimodal diffusion. In: CVPR. pp. 27154–27163 (2024)

    2024

  31. [31]

    In: ICCV

    Mallis, D., Karadeniz, A.S., Cavada, S., Rukhovich, D., Foteinopoulou, N., Cherenkova, K., Kacem, A., Aouada, D.: CAD-Assistant: tool-augmented vllms as generic cad task solvers. In: ICCV. pp. 7284–7294 (2025)

    2025

  32. [32]

    In: ECCV (2020)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: NeRF: Representing scenes as neural radiance fields for view synthesis. In: ECCV (2020)

    2020

  33. [33]

    arXiv preprint arXiv:2506.00568 (2025)

    Niu,K.,Chen,Z.,Yu,H.,Chen,Y.,Fu,T.,Zhao,M.,Li,B.,Xue,X.:CReFT-CAD: Boosting orthographic projection reasoning for cad via reinforcement fine-tuning. arXiv preprint arXiv:2506.00568 (2025)

    work page arXiv 2025

  34. [34]

    arXiv preprint arXiv:2508.10118 (2025)

    Niu, K., Yu, H., Chen, Z., Zhao, M., Fu, T., Li, B., Xue, X.: From intent to execution: Multimodal chain-of-thought reinforcement learning for precise cad code generation. arXiv preprint arXiv:2508.10118 (2025)

    work page arXiv 2025

  35. [35]

    Qi et al

    OpenAI: Introducing gpt 5.2 (2025),https://openai.com/index/introducing- gpt-5-2 18 D. Qi et al

    2025

  36. [36]

    In: CVPR

    Qi, D., Wang, C., Xu, J., Chu, T., Zhao, Z., Liu, W., Ding, W., Ma, Y., Gao, S.: Pointer-CAD: Unifying b-rep and command sequences via pointer-based edges & faces selection. In: CVPR. pp. 17377–17387 (2026)

    2026

  37. [37]

    In: ACM MM

    Qin, F., Lu, S., Hou, J., Wang, C., Fang, M., Liu, L.: Drawing2CAD: Sequence- to-sequence learning for cad generation from vector drawings. In: ACM MM. pp. 10573–10582 (2025)

    2025

  38. [38]

    In: ICCV

    Rukhovich, D., Dupont, E., Mallis, D., Cherenkova, K., Kacem, A., Aouada, D.: CAD-Recode: Reverse engineering cad code from point clouds. In: ICCV. pp. 9801– 9811 (2025)

    2025

  39. [39]

    Proceedings of the Design Society5, 1893–1902 (2025)

    Schüpbach, A., San Miguel, R., Ferchow, J., Meboldt, M.: From text to design: a framework to leverage llm agents for automated cad generation. Proceedings of the Design Society5, 1893–1902 (2025)

    1902

  40. [40]

    Neurocomputing568, 127063 (2024)

    Su, J., Ahmed, M., Lu, Y., Pan, S., Bo, W., Liu, Y.: RoFormer: Enhanced trans- former with rotary position embedding. Neurocomputing568, 127063 (2024)

    2024

  41. [41]

    NeurIPS (2020)

    Tancik, M., Srinivasan, P.P., Mildenhall, B., Fridovich-Keil, S., Raghavan, N., Sing- hal, U., Ramamoorthi, R., Barron, J.T., Ng, R.: Fourier features let networks learn high frequency functions in low dimensional domains. NeurIPS (2020)

    2020

  42. [42]

    Kimi K2.5: Visual Agentic Intelligence

    Team, K., Bai, T., Bai, Y., Bao, Y., Cai, S., Cao, Y., Charles, Y., Che, H., Chen, C., Chen, G., et al.: Kimi K2. 5: Visual agentic intelligence. arXiv preprint arXiv:2602.02276 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  43. [43]

    In: Proceedings of the 17th ACM Conference on Recommender Systems

    Wang, C., Ye, J., Wang, W., Gao, C., Feng, F., He, X.: RecAD: Towards a unified library for recommender attack and defense. In: Proceedings of the 17th ACM Conference on Recommender Systems. pp. 234–244 (2023)

    2023

  44. [44]

    In: AAAI

    Wang, S., Chen, C., Le, X., Xu, Q., Xu, L., Zhang, Y., Yang, J.: CAD-GPT: Syn- thesising cad construction sequence with spatial reasoning-enhanced multimodal llms. In: AAAI. vol. 39, pp. 7880–7888 (2025)

    2025

  45. [45]

    ACM TOG40(4), 1–24 (2021)

    Willis, K.D., Pu, Y., Luo, J., Chu, H., Du, T., Lambourne, J.G., Solar-Lezama, A., Matusik, W.: Fusion 360 gallery: A dataset and environment for programmatic cad construction from human design sequences. ACM TOG40(4), 1–24 (2021)

    2021

  46. [46]

    In: ICCV

    Wu, R., Xiao, C., Zheng, C.: DeepCAD: A deep generative network for computer- aided design models. In: ICCV. pp. 6772–6782 (2021)

    2021

  47. [47]

    In: ECCV

    Wu, S., Khasahmadi, A.H., Katz, M., Jayaraman, P.K., Pu, Y., Willis, K., Liu, B.: CadVLM: Bridging language and vision in the generation of parametric cad sketches. In: ECCV. pp. 368–384. Springer (2024)

    2024

  48. [48]

    Text-to-cadquery: A new paradigm for cad generation with scalable large model capabilities.arXiv preprint arXiv:2505.06507, 2025

    Xie, H., Ju, F.: Text-to-CadQuery: A new paradigm for cad generation with scal- able large model capabilities. arXiv preprint arXiv:2505.06507 (2025)

    work page arXiv 2025

  49. [49]

    Cad-mllm: Unify- ing multimodality-conditioned cad generation with mllm.arXiv preprint arXiv:2411.04954, 2024

    Xu, J., Wang, C., Zhao, Z., Liu, W., Ma, Y., Gao, S.: CAD-MLLM: Uni- fying multimodality-conditioned cad generation with mllm. arXiv preprint arXiv:2411.04954 (2024)

    work page arXiv 2024

  50. [50]

    arXiv preprint arXiv:2307.00149 (2023)

    Xu, X., Jayaraman, P.K., Lambourne, J.G., Willis, K.D., Furukawa, Y.: Hi- erarchical neural coding for controllable cad model generation. arXiv preprint arXiv:2307.00149 (2023)

    work page arXiv 2023

  51. [51]

    Skexgen: Autoregressive generation of cad construction sequences with disentangled codebooks

    Xu, X., Willis, K.D., Lambourne, J.G., Cheng, C.Y., Jayaraman, P.K., Furukawa, Y.: SkexGen: Autoregressive generation of cad construction sequences with disen- tangled codebooks. arXiv preprint arXiv:2207.04632 (2022)

    work page arXiv 2022

  52. [52]

    Qwen3 Technical Report

    Yang, A., Li, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Gao, C., Huang, C., Lv, C., et al.: Qwen3 technical report. arXiv preprint arXiv:2505.09388 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [53]

    Computer-Aided Design p

    Yin, X., Lu, X., Shen, J., Ni, J., Li, H., Tong, R., Tang, M., Du, P.: RLCAD: Re- inforcement learning training gym for revolution involved cad command sequence generation. Computer-Aided Design p. 104027 (2025) Pointer-CAD v2 19

    2025

  54. [54]

    In: Proceedings of the SIGGRAPH Asia 2025 Conference

    You, Y., Uy, M.A., Han, J., Thomas, R., Zhang, H., Du, Y., Chen, H., Engelmann, F., You, S., Guibas, L.: Img2CAD: Reverse engineering 3d cad models from images through vlm-assisted conditional factorization. In: Proceedings of the SIGGRAPH Asia 2025 Conference. pp. 1–12 (2025)

    2025

  55. [55]

    Journal of Mechanical Design148(3), 031703 (2026)

    Yu, N., Ferdous Alam, M., Hart, A.J., Ahmed, F.: GenCAD-three-dimensional: Computer-aided design program generation using multimodal latent space align- ment and synthetic dataset balancing. Journal of Mechanical Design148(3), 031703 (2026)

    2026

  56. [56]

    Computers & Graphics124, 104048 (2024)

    Yuan, Z., Shi, J., Huang, Y.: OpenECAD: An efficient visual language model for editable 3d-cad design. Computers & Graphics124, 104048 (2024)

    2024

  57. [57]

    IEEE TVCG 31(12), 10188–10199 (2025)

    Zhang, A., Jia, W., Zou, Q., Feng, Y., Wei, X., Zhang, Y.: Diffusion-CAD: Control- lable diffusion model for generating computer-aided design models. IEEE TVCG 31(12), 10188–10199 (2025)

    2025

  58. [58]

    Computer-Aided Design178, 103806 (2025)

    Zhang, C., Polette, A., Pinquié, R., Carasi, G., De Charnace, H., Pernot, J.P.: eCAD-Net: Editable parametric cad models reconstruction from dumb b-rep mod- els using deep neural networks. Computer-Aided Design178, 103806 (2025)

    2025

  59. [59]

    In: Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining

    Zheng,Q.,Xia,X.,Zou,X.,Dong,Y.,Wang,S.,Xue,Y.,Shen,L.,Wang,Z.,Wang, A., Li, Y., et al.: CodeGeeX: A pre-trained model for code generation with multi- lingual benchmarking on humaneval-x. In: Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. pp. 5673–5684 (2023)

    2023

  60. [60]

    In: ACM MM

    Zheng, W., Sun, C., Zhang, W., Lv, J., Liu, X.: Target-guided bayesian flow net- works for quantitatively constrained cad generation. In: ACM MM. pp. 3330–3339 (2025)

    2025

  61. [61]

    Computer-Aided De- sign p

    Zhou, J., Camba, J.D., Company, P.: CADialogue: A multimodal llm-powered con- versational assistant for intuitive parametric cad modeling. Computer-Aided De- sign p. 104006 (2025) 20 D. Qi et al. Supplementary Material In these supplementary materials, we provide the following: –Additional experiments and analysis, including further comparisons on the Omn...

    work page arXiv 2025

  62. [62]

    The sketch plane definition (origin and rotation)

  63. [63]

    The sketch geometry — parameters of all lines, arcs, and circles used in the profile

  64. [64]

    the vertical outer edge formed by the last extrusion

    The extrude operation parameters (normal and reverse-normal lengths). Do not describe assemblies, boolean operations, fillets, chamfers, or any earlier sketches or solids. General Rules –Every numeric parameter (length, angle, radius, coordinate, etc.) must be defined with a tag:<v1=10mm>,<v2=0.5m>,<v3=30deg>, etc. –All values must use consistent units. •...