pith. sign in

arxiv: 2602.19171 · v2 · submitted 2025-12-08 · 💻 cs.GR · cs.AI

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

Xintong Dong , Chuanyang Li , Peng Zheng , Chuqi Han , Jiaxin Jing , Hailong Shen , Yanzhi Song , Zhouwang Yang This is my paper

Pith reviewed 2026-05-17 01:10 UTC · model grok-4.3

classification 💻 cs.GR cs.AI
keywords parametric CADdesign intentgeometric constraintsCAD benchmarkparametric sequenceseditabilityCAD generationindustrial dataset
0
0 comments X

The pith

Explicit constraints in parametric CAD sequences preserve design intent after parameter edits.

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

The paper introduces HistCAD as a software-independent representation for parametric CAD histories that records sketch primitives, dimensional and geometric constraints, feature operations, and boundary references. It supplies a dataset of 170,236 executable sequences matched to native industrial CAD models, STEP files, renders, and text descriptions. A new Constraint-Aware Editability Benchmark applies edits and scores results with Edit Reachability, conditional preserved constraint satisfaction, and Overall Editable Success. Experiments establish that models without explicit constraints lose design intent far more often than constraint-aware ones, and that the representation supports text-supervised generation plus direct LLM use. This shifts the goal of CAD generation from matching static shapes to producing reusable, editable sequences.

Core claim

HistCAD defines an intermediate language that records sketch primitives, constraints, feature operations, and 3D point boundary references for operations such as fillet and chamfer. The accompanying dataset of 170,236 sequences is aligned with native CAD models and STEP files. The Constraint-Aware Editability Benchmark measures Edit Reachability, conditional preserved constraint satisfaction, and Overall Editable Success to separate failures to reach a valid state from failures to keep required constraints. Experiments show that explicit constraints are essential for preserving design intent after edits and that HistCAD enables supervised CAD generation from text and direct LLM workflows.

What carries the argument

HistCAD intermediate language that records constraints and operations independently of CAD software to measure and enforce editability of parametric sequences.

If this is right

  • CAD generation systems can be trained to output constraint-explicit sequences that remain valid after parameter changes.
  • Text-to-CAD pipelines gain practical utility because generated models preserve intended behavior under edits.
  • LLM-based CAD workflows can produce reusable histories directly rather than static geometry.
  • Evaluation of new CAD generators can now distinguish reachability errors from constraint-violation errors.

Where Pith is reading between the lines

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

  • The representation could support automatic conversion of legacy CAD files into editable constraint-aware forms.
  • Training on the 170k-scale dataset may improve generalization of generative models across different industrial domains.
  • Extending the benchmark to assembly-level constraints or multi-step feature edits would test broader applicability.

Load-bearing premise

The introduced metrics of Edit Reachability, conditional preserved constraint satisfaction, and Overall Editable Success, together with dataset alignment to native CAD models, sufficiently capture real-world design intent preservation and industrial complexity.

What would settle it

Running the edit benchmark on CAD sequences generated by methods that omit explicit constraints and finding that their Overall Editable Success rates equal or exceed those of HistCAD-trained models would falsify the claim that explicit constraints are essential.

Figures

Figures reproduced from arXiv: 2602.19171 by Chuanyang Li, Chuqi Han, Hailong Shen, Jiaxin Jing, Peng Zheng, Xintong Dong, Yanzhi Song, Zhouwang Yang.

Figure 1
Figure 1. Figure 1: Left: A diverse sample of industrial CAD models from the HistCAD dataset. Right: A local edit example illustrating the necessity of explicit constraints. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between hierarchical sketch serialization and HistCAD’s flat sketch representation. Left: hierarchical serialization repeats loop membership [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Protocol and case partition for the Constraint-Aware Editability Benchmark. The benchmark applies the target dimensional edit to each evaluated [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Token count distributions across the 134,896-model shared intersec [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative samples from the HistCAD-DeepCAD, HistCAD-Fusion360, and HistCAD-Industrial subsets. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustrative examples from the native sequence constraint ablation study. For each reference sequence, we compare a variant with full constraints [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Direct HistCAD sequence generation with GPT-5.5. Given the HistCAD specification and design requirements in natural language, GPT-5.5 outputs [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: HistCAD JSON sequences generated directly by GPT-5.5 for Figure [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
read the original abstract

Parametric CAD sequences are reusable because dimensional and geometric constraints govern how parameter changes propagate. Existing CAD generation datasets and benchmarks emphasize reconstruction fidelity, execution validity, or static shape similarity, leaving preservation of design intent under edits largely unmeasured. We introduce HistCAD, a representation standard, dataset, and benchmark for executable parametric CAD with explicit constraints. HistCAD defines an intermediate language independent of CAD software, recording sketch primitives, constraints, feature operations, and 3D point boundary references for operations such as fillet and chamfer. The dataset contains 170,236 executable sequences aligned with native CAD models, STEP files, rendered views, and text annotations, combining academic scale with professionally authored industrial complexity. Building on this representation, the Constraint-Aware Editability Benchmark applies parameter edits and reports Edit Reachability, conditional preserved constraint satisfaction, and Overall Editable Success, abbreviated ER, cPCSR, and OES; these metrics separate failures to reach a valid edited state from failures to preserve required constraints. Experiments show that explicit constraints are essential for preserving design intent after edits, and that HistCAD supports supervised CAD generation from text and direct LLM workflows. We argue that HistCAD reframes CAD generation from static shape imitation to the synthesis of reusable parametric sequences with explicit constraints.

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

3 major / 2 minor

Summary. The paper introduces HistCAD, a constraint-aware parametric history-based CAD representation using an intermediate language that records sketch primitives, explicit constraints, feature operations, and 3D boundary references. It releases a dataset of 170,236 executable sequences aligned with native CAD models, STEP files, rendered views, and text annotations that combines academic scale with industrial complexity. A new Constraint-Aware Editability Benchmark is defined with metrics Edit Reachability (ER), conditional preserved constraint satisfaction (cPCSR), and Overall Editable Success (OES) that separate failures to reach a valid edited state from failures to preserve constraints. Experiments are reported to show that explicit constraints are essential for preserving design intent after edits and that HistCAD enables supervised CAD generation from text as well as direct LLM workflows.

Significance. If the results hold, HistCAD could meaningfully shift CAD generation research from static shape imitation toward synthesis of reusable parametric sequences that preserve design intent under editing. The dataset scale, alignment with professionally authored industrial models, and the new metrics that disentangle reachability from constraint preservation are concrete strengths that could support more reliable generative CAD pipelines and LLM integration. Credit is due for releasing an executable representation standard and benchmark rather than another reconstruction-only corpus.

major comments (3)
  1. [Dataset construction] Dataset construction section: alignment of HistCAD sequences with native CAD models and STEP files is described, but no expert validation or analysis is reported to confirm that the extracted explicit constraints capture relevant implicit relations or manufacturing constraints typical in industrial practice. This is load-bearing for the claim that higher cPCSR/OES scores demonstrate preservation of design intent rather than merely improved executability.
  2. [Experiments] Experiments section: the claim that experiments demonstrate the value of explicit constraints for intent preservation is stated, yet details on validation procedures, error analysis, and automatic verification of constraint satisfaction are insufficient. Without these, the empirical support for the central necessity argument remains only moderately grounded.
  3. [Constraint-Aware Editability Benchmark] Benchmark definition: ER, cPCSR, and OES treat the constraints recorded in HistCAD as ground-truth intent. If the dataset alignment misses implicit design relations, superior scores for constraint-aware models would indicate only better executability, weakening the argument that explicit constraints are essential for real-world design intent preservation.
minor comments (2)
  1. [Notation and presentation] Ensure all abbreviations (ER, cPCSR, OES) are defined at first use and used consistently in text, tables, and figure captions.
  2. [Figures] Figure captions for rendered views and edit examples should explicitly note whether the displayed models satisfy the reported constraint sets.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below. Revisions have been made to improve clarity, add details, and acknowledge limitations where appropriate.

read point-by-point responses

  1. Referee: [Dataset construction] Dataset construction section: alignment of HistCAD sequences with native CAD models and STEP files is described, but no expert validation or analysis is reported to confirm that the extracted explicit constraints capture relevant implicit relations or manufacturing constraints typical in industrial practice. This is load-bearing for the claim that higher cPCSR/OES scores demonstrate preservation of design intent rather than merely improved executability.

    Authors: We agree that expert validation would provide additional support. HistCAD sequences are extracted directly from the parametric histories of the source industrial CAD models, so the recorded constraints are precisely the explicit ones used by the original designers. Alignment is verified by successful execution to matching STEP geometry and rendered views. We acknowledge that unrecorded implicit relations or manufacturing constraints may exist outside the history. The revised manuscript adds a dedicated limitations paragraph clarifying that cPCSR/OES evaluate preservation of the recorded explicit constraints and discusses the distinction from implicit intent. revision: partial

  2. Referee: [Experiments] Experiments section: the claim that experiments demonstrate the value of explicit constraints for intent preservation is stated, yet details on validation procedures, error analysis, and automatic verification of constraint satisfaction are insufficient. Without these, the empirical support for the central necessity argument remains only moderately grounded.

    Authors: We appreciate the request for greater rigor. The original text summarized the setup at a high level. The revised Experiments section now includes: (1) explicit description of the automatic constraint-satisfaction checker implemented via the CAD kernel API, (2) breakdown of failure modes with quantitative error analysis across the test set, and (3) additional ablation tables isolating the contribution of explicit constraints. These additions directly strengthen the empirical grounding of the necessity claim. revision: yes

  3. Referee: [Constraint-Aware Editability Benchmark] Benchmark definition: ER, cPCSR, and OES treat the constraints recorded in HistCAD as ground-truth intent. If the dataset alignment misses implicit design relations, superior scores for constraint-aware models would indicate only better executability, weakening the argument that explicit constraints are essential for real-world design intent preservation.

    Authors: This is a fair observation. The benchmark is deliberately defined relative to the constraints present in the aligned HistCAD sequences. Higher cPCSR and OES therefore demonstrate superior preservation of those recorded constraints under editing. We have revised the benchmark section to state this scope explicitly and to note that the metrics do not claim to capture unrecorded implicit relations. The distinction is now discussed in both the benchmark definition and the limitations paragraph. revision: yes

standing simulated objections not resolved
  • Absence of independent expert validation confirming that extracted constraints capture implicit manufacturing relations typical in industrial practice

Circularity Check

0 steps flagged

No significant circularity: new representation, dataset, and benchmark with explicitly defined metrics

full rationale

The paper introduces HistCAD as a new intermediate language and representation for parametric CAD sequences that include explicit constraints, along with a dataset of 170,236 sequences and a benchmark using newly defined metrics (ER, cPCSR, OES). These metrics are constructed directly from the HistCAD format to measure edit reachability and constraint preservation, but the central claims rest on empirical evaluation against this new resource rather than any closed derivation or self-referential equations. No load-bearing self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work are evident in the provided abstract and context; the contribution is constructive and self-contained against external CAD models and STEP files.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The work rests on the domain assumption that parametric CAD models are defined by reusable constraints and history; the central addition is the new representation and data rather than fitted parameters or new physical entities.

axioms (1)
  • domain assumption Parametric CAD sequences are reusable because dimensional and geometric constraints govern how parameter changes propagate.
    Directly stated in the abstract as the foundation for the representation.
invented entities (1)
  • HistCAD intermediate language no independent evidence
    purpose: Records sketch primitives, constraints, feature operations, and 3D point boundary references independent of specific CAD software.
    Newly defined construct introduced by the paper to enable constraint-aware history tracking.

pith-pipeline@v0.9.0 · 5550 in / 1236 out tokens · 37534 ms · 2026-05-17T01:10:27.769865+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

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

  1. BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CAD

    cs.AI 2026-05 unverdicted novelty 7.0

    BenchCAD is a new benchmark showing that frontier multimodal models recover coarse geometry but fail to generate faithful parametric CAD programs for industrial parts.

  2. BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CAD

    cs.AI 2026-05 unverdicted novelty 7.0

    BenchCAD benchmark shows frontier multimodal models recover coarse geometry but fail to produce accurate parametric CAD programs for industrial parts, with limited generalization after fine-tuning.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    https://www.heldermann-verlag.de/jgg/ jgg03/jgg0303.htm Autodesk

    Using geometric constraints to capture design intent.Journal for Geometry and Graphics3, 1 (1999), 39–45. https://www.heldermann-verlag.de/jgg/ jgg03/jgg0303.htm Autodesk. 2014.Fusion 360 API. http://help.autodesk.com/view/fusion360/ENU/?guid= GUID-7B5A90C8-E94C-48DA-B16B-430729B734DC Bernhard Bettig and Christoph M. Hoffmann

    work page 1999

  2. [2]

    doi:10.1115/1.3593408 Jorge D

    Geometric constraint solving in parametric computer-aided design.Journal of Computing and Information Science in Engineering11, 2 (2011), 021001. doi:10.1115/1.3593408 Jorge D. Camba, Manuel Contero, and Pedro Company

    work page doi:10.1115/1.3593408 2011

  3. [3]

    and Contero, Manuel and Company, Pedro , date =

    Parametric CAD modeling: An analysis of strategies for design reusability.Computer-Aided Design74 (2016), 18–31. doi:10.1016/j.cad.2016.01.003 Tianrun Chen, Chunan Yu, Yuanqi Hu, Jing Li, Tao Xu, Runlong Cao, Lanyun Zhu, Ying Zang, Yong Zhang, Zejian Li, et al

    work page doi:10.1016/j.cad.2016.01.003 2016

  4. [4]

    doi:10.1109/TII.2025.3584476 Pedro Company, Ferran Naya, Manuel Contero, and Jorge D

    Img2CAD: Conditioned 3-D CAD Model Generation From Single Image With Structured Visual Geometry.IEEE Transactions on Industrial Informatics21, 11 (2025), 8539–8549. doi:10.1109/TII.2025.3584476 Pedro Company, Ferran Naya, Manuel Contero, and Jorge D. Camba

    work page doi:10.1109/tii.2025.3584476 2025

  5. [5]

    On the role of geometric constraints to support design intent communication and model reusability.Computer-Aided Design and Applications17, 1 (2020), 61–76. doi:10. 14733/cadaps.2020.61-76 CadQuery contributors. 2026.CadQuery. doi:10.5281/zenodo.18633916 Elona Dupont, Kseniya Cherenkova, Dimitrios Mallis, Gleb Gusev, Anis Kacem, and Djamila Aouada

    work page doi:10.5281/zenodo.18633916 2020

  6. [6]

    InEuropean Conference on Computer Vision (ECCV)

    TransCAD: A hierarchical transformer for CAD sequence inference from point clouds. InEuropean Conference on Computer Vision (ECCV). 19–36. doi:10.1007/978-3-031-73030-6_2 Haoqiang Fan, Hao Su, and Leonidas J. Guibas

    work page doi:10.1007/978-3-031-73030-6_2

  7. [7]

    InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    A Point Set Generation Network for 3D Object Reconstruction from a Single Image. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 605–613. Rubin Fan, Fazhi He, Yuxin Liu, and Jing Lin. 2025a. A history-based parametric CAD sketch dataset with advanced engineering commands.Computer-Aided Design182 (2025), 103848. doi:...

    work page doi:10.1016/j.cad.2025.103848 2025

  8. [8]

    https: //openreview.net/forum?id=lExqWvQht8 Yandong Guan, Xilin Wang, Ximing Xing, Jing Zhang, Dong Xu, and Qian Yu

    CADmium: Fine-Tuning Code Language Models for Text-Driven Sequential CAD Design.Transactions on Machine Learning Research(2026). https: //openreview.net/forum?id=lExqWvQht8 Yandong Guan, Xilin Wang, Ximing Xing, Jing Zhang, Dong Xu, and Qian Yu

    work page 2026

  9. [9]

    ACM Transactions on Graphics , author =

    Com- plexGen: CAD reconstruction by B-rep chain complex generation.ACM Transactions on Graphics (TOG)41, 4, Article 129 (2022), 18 pages. doi:10.1145/3528223.3530078 Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    work page doi:10.1145/3528223.3530078 2022

  10. [10]

    https: //openreview.net/forum?id=ZR2CDgADRo JiuShao

    SolidGen: An autoregressive model for direct B-rep synthesis.Transactions on Machine Learning Research(2023). https: //openreview.net/forum?id=ZR2CDgADRo JiuShao. 2025.JiuShao POWER Platform API. https://docs.amcax.net/power/v2_5_0/ zh_cn/html/index.html Mohammad Sadil Khan, Elona Dupont, Sk Aziz Ali, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. 20...

    work page 2023

  11. [11]

    Towards High-Fidelity CAD Generation via LLM-Driven Program Generation and Text-Based B-Rep Primitive Grounding

    CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 18563–18573. Jiahao Li, Qingwang Zhang, Qiuyu Chen, Guozhan Qiu, Yunzhong Lou, and Xiangdong Zhou. 2026b. Towards High-Fidelity CAD Generation via LLM-Driven Progr...

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    doi:10.1016/0010- 4485(82)90292-5 Dawei Lin, Meng Yuan, Ziming Wang, Tieru Wu, and Yuanning Liu

    Modification of geometric models through varia- tional geometry.Computer-Aided Design14, 4 (1982), 209–214. doi:10.1016/0010- 4485(82)90292-5 Dawei Lin, Meng Yuan, Ziming Wang, Tieru Wu, and Yuanning Liu

    work page doi:10.1016/0010- 1982

  13. [13]

    InProceedings of the 33rd ACM International Conference on Multimedia (ACM MM)

    FreeCAD: A Multimodal Framework for 3D CAD Model Generation from Free-Form Prompts. InProceedings of the 33rd ACM International Conference on Multimedia (ACM MM). 1948–1956. doi:10.1145/3746027.3755550 Yilin Liu, Niladri Shekhar Dutt, Changjian Li, and Niloy J. Mitra

    work page doi:10.1145/3746027.3755550 1948

  14. [14]

    Weijian Ma, Shuaiqi Chen, Yunzhong Lou, Xueyang Li, and Xiangdong Zhou

    B-repLer: Seman- tic B-rep Latent Editor using Large Language Models.arXiv preprint arXiv:2508.10201 (2025). Weijian Ma, Shuaiqi Chen, Yunzhong Lou, Xueyang Li, and Xiangdong Zhou

    work page arXiv 2025

  15. [15]

    InProceedings of the 33rd ACM International Conference on Multimedia (ACM MM)

    Drawing2CAD: Sequence-to-Sequence Learning for CAD Generation from Vector Drawings. InProceedings of the 33rd ACM International Conference on Multimedia (ACM MM). 10573–10582. doi:10.1145/3746027.3755782 Feiwei Qin, Chenqi Luo, Junhao Hou, Meie Fang, and Ligang Liu

    work page doi:10.1145/3746027.3755782

  16. [16]

    BRep-GD: A Graph Diffusion Model for CAD Boundary Representation Generation.IEEE Transactions on Visualization and Computer Graphics(2026), 1–17. doi:10.1109/ TVCG.2026.3676456 HistCAD: A Constraint-Aware Parametric History-Based CAD Representation, Dataset, and Benchmark with Industrial Complexity•13 Danila Rukhovich, Elona Dupont, Dimitrios Mallis, Ksen...

    work page arXiv 2026

  17. [17]

    InICML 2020 Workshop on Object-Oriented Learning

    SketchGraphs: A Large-Scale Dataset for Modeling Relational Geometry in Computer-Aided Design. InICML 2020 Workshop on Object-Oriented Learning. Yue Sun, Jituo Li, Ziqin Xu, Jialu Zhang, Xinqi Liu, Dongliang Zhang, and Guodong Lu

    work page 2020

  18. [18]

    doi:10.1109/TVCG.2025.3566544 Ruiyu Wang, Yu Yuan, Shizhao Sun, and Jiang Bian

    Sketch2Seq: Reconstruct CAD Models From Feature-Based Sketch Seg- mentation.IEEE Transactions on Visualization and Computer Graphics31, 10 (2025), 8214–8230. doi:10.1109/TVCG.2025.3566544 Ruiyu Wang, Yu Yuan, Shizhao Sun, and Jiang Bian. 2025b. Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models. InInternational Conference on ...

    work page doi:10.1109/tvcg.2025.3566544 2025

  19. [19]

    doi:10.1145/3450626.3459818 Rundi Wu, Chang Xiao, and Changxi Zheng

    Fusion 360 Gallery: A Dataset and Environment for Programmatic CAD Construction from Human Design Se- quences.ACM Transactions on Graphics (TOG)40, 4, Article 54 (2021), 24 pages. doi:10.1145/3450626.3459818 Rundi Wu, Chang Xiao, and Changxi Zheng

    work page doi:10.1145/3450626.3459818 2021

  20. [20]

    Text-to-CadQuery: A New Paradigm for CADgenerationwithscalablelargemodelcapabilities

    Text-to-CadQuery: A New Paradigm for CAD Generation with Scalable Large Model Capabilities. arXiv:2505.06507 [cs.AI] https://arxiv.org/abs/2505.06507 Xiang Xu, Pradeep Kumar Jayaraman, Joseph G Lambourne, Karl DD Willis, and Yasutaka Furukawa

    work page arXiv

  21. [21]

    doi:10.1145/3658129 Xiang Xu, Karl DD Willis, Joseph G Lambourne, Chin-Yi Cheng, Pradeep Kumar Jayara- man, and Yasutaka Furukawa

    BrepGen: A B-rep Generative Diffusion Model with Structured Latent Geometry.ACM Transactions on Graphics (TOG)43, 4, Article 119 (2024), 14 pages. doi:10.1145/3658129 Xiang Xu, Karl DD Willis, Joseph G Lambourne, Chin-Yi Cheng, Pradeep Kumar Jayara- man, and Yasutaka Furukawa

    work page doi:10.1145/3658129 2024

  22. [22]

    Qwen3 Technical Report

    Qwen3 Technical Report.arXiv preprint arXiv:2505.09388(2025). Yang You, Mikaela Angelina Uy, Jiaqi Han, Rahul Thomas, Haotong Zhang, Yi Du, Han- sheng Chen, Francis Engelmann, Suya You, and Leonidas Guibas

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    contained

    Img2CAD: Reverse engineering 3D CAD models from images through VLM-assisted condi- tional factorization. InSIGGRAPH Asia 2025 Conference Papers. 1–12. doi:10.1145/ 3757377.3763891 Chao Zhang, Arnaud Polette, Romain Pinquié, Gregorio Carasi, Henri De Charnace, and Jean-Philippe Pernot. 2025a. eCAD-Net: Editable Parametric CAD Models Reconstruction from Dum...

    work page doi:10.1016/j.cad.2024.103806 2025