pith. machine review for the scientific record. sign in

arxiv: 2605.13293 · v1 · submitted 2026-05-13 · 💻 cs.CV

Recognition: unknown

Img2CADSeq: Image-to-CAD Generation via Sequence-Based Diffusion

Shiyu Tan , Zixuan Zhao , Hao Gao , Zhiheng Chen , Xiaolong Yin , Enya Shen
Authors on Pith no claims yet

Pith reviewed 2026-05-14 19:53 UTC · model grok-4.3

classification 💻 cs.CV
keywords image-to-CADCAD sequence generationdiffusion modelsBRep reconstructionVQ-Diffusionpoint cloud conditioninghierarchical codebook
0
0 comments X

The pith

Img2CADSeq generates valid CAD models from single images by compressing operation sequences into a hierarchical codebook and diffusing them from a point-cloud bridge.

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

The paper presents a pipeline that converts single-view images into complete CAD models in the standard STEP format used by commercial software. It tackles the difficulty of preserving topological rules and operation order by first turning long CAD sequences into a compact three-level codebook that prioritizes profiles and key features. A coarse-to-fine point cloud serves as an intermediate representation that aligns image features with 3D sequences through contrastive learning, allowing a VQ-Diffusion model to produce realistic sequences. New datasets CAD-220K and PrintCAD provide the scale needed for training across industrial cases. Experiments show the resulting models outperform prior methods and load directly into existing CAD tools.

Core claim

Encoding CAD sequences into a three-level hierarchical codebook guided by importance prioritization compresses long sequences into a stable discrete latent space that preserves topological constraints and operation validity, enabling a contrastive point-cloud bridge from 2D images to condition a VQ-Diffusion model that outputs usable CAD sequences.

What carries the argument

Three-level hierarchical codebook for CAD sequences that compresses long operation lists while keeping topological validity and prioritizing profiles over secondary details.

If this is right

  • Generated models appear as standard STEP files that load and edit without conversion in tools such as SolidWorks or Fusion 360.
  • The method supports direct industrial use because the new CAD-220K and PrintCAD datasets improve adaptation to real manufacturing data.
  • Sequence generation respects operation order and validity, reducing the need for post-processing cleanup.
  • Single-image input becomes sufficient for high-quality reconstruction where earlier methods required multiple views or manual intervention.

Where Pith is reading between the lines

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

  • The same hierarchical encoding might apply to other ordered design domains such as circuit netlists or mechanical assembly plans.
  • Extending the point-cloud bridge to multi-view images could further reduce ambiguity in complex topologies.
  • Integration with existing CAD version-control systems could allow automatic reconstruction of legacy parts from archived photographs.

Load-bearing premise

Encoding CAD sequences into a three-level hierarchical codebook guided by importance prioritization can compress long sequences into a stable discrete latent space while preserving the topological constraints and operation validity required for usable CAD models.

What would settle it

Generate STEP files from test images and attempt to open and edit them in commercial CAD software; failure to load without errors or violation of geometric constraints would falsify the claim.

Figures

Figures reproduced from arXiv: 2605.13293 by Enya Shen, Hao Gao, Shiyu Tan, Xiaolong Yin, Zhiheng Chen, Zixuan Zhao.

Figure 1
Figure 1. Figure 1: Our proposed Img2CADSeq is a novel method based on boundary representations (BReps) structure. It is a multi-stage pipeline that can generate standardized STEP files. The fourth and fifth columns show reconstructed results generated with single-view image conditioning. The method also delivers strong results in unconditional generation, as seen in the first three columns with red parts and cloud-conditione… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Img2CADSeq Framework. In the first stage, hierarchical sequence encoding represents CAD operations via a three-level codebook into a discrete space. Then we lift the input image into a 3D point cloud using a tailored network trained jointly on both synthetic and real-world data types, which is then refined by UA-DGCNN to sharpen edges and smooth surfaces. Finally, we employ contrastive lear… view at source ↗
Figure 3
Figure 3. Figure 3: Workflow of Hierarchical Entity Construction. At the base level, the Curve-Cluster parameterizes geometric primitives, which form closed loops in the Sketch-Patch. These loops are then lifted into 3D space via a normal vector and origin to perform extrusion and Boolean operations, resulting in an Extrude-Block. Multiple blocks are finally assembled to yield the target solid. This process mirrors the constr… view at source ↗
Figure 4
Figure 4. Figure 4: The limitations and failure cases of our work. (a) Extreme single-view ambiguity causes plausible back-end structures but non￾manufacturable in occluded areas. (b)Sequence error accumulation disrupts global geometric constraints like strict symmetry and coaxiality. (c) Lim￾ited resolution of the intermediate point cloud causes fine features to be smoothed out or omitted. method’s leading performance in Tab… view at source ↗
Figure 5
Figure 5. Figure 5: Here are some samples from our newly introduced dataset, PrintCAD, which comprises over 2,000 3D printed objects captured under uncontrolled [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: We evaluate our method on synthetic and challenging real-world images. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: We evaluate our method against state-of-the-art approaches on inputs with ill-scanned point clouds with misalignment parts, or the clean ones. While [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: We compare our Img2CADSeq with other widely adopted baselines in unconditional generation. Our method produces structurally plausible models [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Current CAD datasets, such as DeepCAD, predominantly [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Random test set samples. While simple geometries are easily handled by most methods, increasing complexity challenges all approaches. Despite limitations in extreme cases, our method better preserves global shape and visual consistency, demonstrating robustness without selection bias. Input Model 4 Full Model GT Input Model 4 Full Model GT [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visual ablation study on fine-tuning with the two datasets. Model 4 exhibits severe geometric distortions and missing features. Our full model captures these complex structural details better, demonstrating the necessity of the extra data for generating industrial-grade CAD models. SIGGRAPH Conference Papers ’26, July 19–23, 2026, Los Angeles, CA, USA [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Reconstruction results from the given images are shown from left to right: input image, BReps, and their CAD vertices and edges. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Reconstruction results from the given images are shown from left to right: input image, BReps, and their CAD vertices and edges. [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
read the original abstract

Boundary Representation (BRep) is the standard format for Computer-Aided Design (CAD), yet reconstructing high-quality BReps from single-view images remains challenging due to the complexity of topological constraints and operation sequences. We present Img2CADSeq, a multi-stage pipeline that overcomes these limitations by encoding CAD sequences into a three-level hierarchical codebook. Guided by an importance prioritization, this strategy values profiles over details, compressing long sequences into a stable discrete latent space. To bridge the modality gap, we leverage a coarse-to-fine point cloud intermediate, aligning 2D visual features with 3D CAD sequences via contrastive learning to condition a VQ-Diffusion model. Supported by newly introduced CAD-220K and PrintCAD datasets, our approach ensures robust industrial domain adaptation. Extensive experiments demonstrate that Img2CADSeq significantly outperforms state-of-the-art methods, producing standard STEP files that can be directly used in commercial CAD software.

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 proposes Img2CADSeq, a multi-stage pipeline for single-view image to CAD sequence generation. It encodes long CAD operation sequences into a three-level hierarchical codebook using importance prioritization (profiles over details), bridges the 2D-3D modality gap via a coarse-to-fine point-cloud intermediate and contrastive learning, and conditions a VQ-Diffusion model to produce sequences that decode to valid BReps exported as STEP files. New datasets CAD-220K and PrintCAD are introduced to support training and industrial domain adaptation, with claims of significant outperformance over prior SOTA methods.

Significance. If the topological validity and direct STEP usability claims hold with quantitative backing, the work would advance image-based CAD reconstruction by addressing sequence length and constraint enforcement through hierarchical discretization, while the new large-scale datasets could serve as community benchmarks for future image-to-CAD research.

major comments (2)
  1. [Experiments] Experiments section: the central claim that decoded sequences are always topologically valid and directly importable as STEP files is not supported by any reported validity metric (e.g., percentage of sequences that parse without self-intersection, invalid extrusion order, or BRep errors); this metric is load-bearing for the 'directly usable in commercial CAD software' assertion.
  2. [Method] §3.2 (hierarchical codebook): the importance-prioritization scheme for the three-level codebook is described at a high level but lacks an explicit proof or empirical demonstration that it preserves operation validity and topological constraints after VQ-Diffusion sampling; a failure rate analysis on decoded sequences is required.
minor comments (2)
  1. [Abstract] Abstract: 'extensive experiments' and 'significantly outperforms' are stated without any numerical results or table references; a one-sentence summary of key metrics would improve readability.
  2. [Method] Notation: the three-level codebook is referred to interchangeably as 'hierarchical' and 'importance-prioritized' without a clear equation defining the prioritization weights or codebook sizes.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comments below regarding the need for quantitative validity metrics and further analysis of the hierarchical codebook. We agree that these additions will strengthen the paper and will incorporate them in the revision.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the central claim that decoded sequences are always topologically valid and directly importable as STEP files is not supported by any reported validity metric (e.g., percentage of sequences that parse without self-intersection, invalid extrusion order, or BRep errors); this metric is load-bearing for the 'directly usable in commercial CAD software' assertion.

    Authors: We acknowledge that the current manuscript does not report a specific quantitative validity metric such as the percentage of decoded sequences that parse successfully without self-intersections, invalid extrusion orders, or BRep errors. While the qualitative results and examples show that generated sequences decode to STEP files importable in commercial CAD software, we agree this metric is important to support the claim. In the revised version, we will add a dedicated validity analysis in the Experiments section, reporting failure rates across the CAD-220K and PrintCAD test sets with details on the validation procedure used. revision: yes

  2. Referee: [Method] §3.2 (hierarchical codebook): the importance-prioritization scheme for the three-level codebook is described at a high level but lacks an explicit proof or empirical demonstration that it preserves operation validity and topological constraints after VQ-Diffusion sampling; a failure rate analysis on decoded sequences is required.

    Authors: The three-level hierarchical codebook with importance prioritization is designed to encode profile operations first to establish core topology before incorporating details, thereby aiming to maintain validity during compression and subsequent sampling. While the manuscript provides a high-level description motivated by CAD semantics, we agree that an empirical demonstration is needed. We will add in the revision a failure rate analysis on decoded sequences, comparing validity rates pre- and post-VQ-Diffusion sampling to show preservation of topological constraints. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper describes a multi-stage pipeline that encodes CAD sequences via a new three-level hierarchical codebook, uses contrastive learning on a coarse-to-fine point cloud bridge, and conditions a standard VQ-Diffusion model. It introduces fresh datasets (CAD-220K, PrintCAD) and reports experimental outperformance on STEP file usability. No step reduces by construction to its own inputs, no fitted parameter is relabeled as a prediction, and no load-bearing claim rests on self-citation chains or imported uniqueness theorems. The central results are externally falsifiable via the reported metrics and commercial CAD import tests.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides insufficient detail to identify specific free parameters, axioms, or invented entities; the hierarchical codebook levels and new datasets are introduced but their exact parameterization and assumptions are not described.

pith-pipeline@v0.9.0 · 5473 in / 1183 out tokens · 65441 ms · 2026-05-14T19:53:12.241062+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

51 extracted references · 16 canonical work pages · 1 internal anchor

  1. [1]

    Communications of the ACM , volume=

    Nerf: Representing scenes as neural radiance fields for view synthesis , author=. Communications of the ACM , volume=. 2021 , publisher=

    2021

  2. [2]

    , author=

    3D Gaussian splatting for real-time radiance field rendering. , author=. ACM Trans. Graph. , volume=

    work page

  3. [3]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Vector quantized diffusion model for text-to-image synthesis , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  4. [4]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Brepnet: A topological message passing system for solid models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  5. [5]

    in 2021 ieee , author=

    Deepcad: A deep generative network for computer-aided design models. in 2021 ieee , author=. CVF International Conference on Computer Vision (ICCV) , pages=

    work page 2021

  6. [6]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Uv-net: Learning from boundary representations , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  7. [7]

    Dens3r: A foundation model for 3d geometry prediction.arXiv preprint arXiv:2507.16290, 2025

    Dens3r: A foundation model for 3d geometry prediction , author=. arXiv preprint arXiv:2507.16290 , year=

    work page arXiv

  8. [8]

    Neurocomputing , volume=

    Roformer: Enhanced transformer with rotary position embedding , author=. Neurocomputing , volume=. 2024 , publisher=

    2024

  9. [9]

    ACM Transactions on Graphics (tog) , volume=

    Dynamic graph cnn for learning on point clouds , author=. ACM Transactions on Graphics (tog) , volume=. 2019 , publisher=

    2019

  10. [10]

    2012 , publisher=

    The NURBS book , author=. 2012 , publisher=

    2012

  11. [11]

    ACM Transactions on Graphics (TOG) , volume=

    BrepGPT: Autoregressive B-rep Generation with Voronoi Half-Patch , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  12. [12]

    Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

    Brepdiff: Single-stage b-rep diffusion model , author=. Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

  13. [13]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Csgnet: Neural shape parser for constructive solid geometry , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  14. [14]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    CAD-Llama: leveraging large language models for computer-aided design parametric 3D model generation , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  15. [15]

    European Conference on Computer Vision , pages=

    Transcad: A hierarchical transformer for cad sequence inference from point clouds , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  16. [16]

    Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    AutoBrep: Autoregressive B-Rep Generation with Unified Topology and Geometry , author=. Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    2025

  17. [17]

    ACM Transactions on Graphics (TOG) , volume=

    Defillet: Detection and removal of fillet regions in polygonal cad models , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  18. [18]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Cad-recode: Reverse engineering cad code from point clouds , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  19. [19]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Wonder3d: Single image to 3d using cross-domain diffusion , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  20. [20]

    arXiv preprint arXiv:2403.02151 , year=

    Triposr: Fast 3d object reconstruction from a single image , author=. arXiv preprint arXiv:2403.02151 , year=

    work page arXiv

  21. [21]

    Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age.arXiv preprint arXiv:2309.03453, 2023

    Syncdreamer: Generating multiview-consistent images from a single-view image , author=. arXiv preprint arXiv:2309.03453 , year=

    work page arXiv

  22. [22]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    CADDreamer: CAD Object Generation from Single-view Images , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  23. [23]

    Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    Img2cad: Reverse engineering 3d cad models from images through vlm-assisted conditional factorization , author=. Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    2025

  24. [24]

    ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    IOVS4NeRF: Incremental Optimal View Selection for Large-Scale NeRFs , author=. ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2025 , organization=

    work page 2025

  25. [25]

    ACM Transactions on Graphics (TOG) , volume=

    Hola: B-rep generation using a holistic latent representation , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  26. [26]

    Proceedings of the ACM on Software Engineering , volume=

    Beyond PEFT: Layer-Wise Optimization for More Effective and Efficient Large Code Model Tuning , author=. Proceedings of the ACM on Software Engineering , volume=. 2025 , publisher=

    2025

  27. [27]

    ShapeNet: An Information-Rich 3D Model Repository

    Shapenet: An information-rich 3d model repository , author=. arXiv preprint arXiv:1512.03012 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    International Conference on Learning Representations (ICLR) 2025 , year =

    FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models , author =. International Conference on Learning Representations (ICLR) 2025 , year =

    2025

  29. [29]

    Advances in neural information processing systems , volume=

    Neural discrete representation learning , author=. Advances in neural information processing systems , volume=

  30. [30]

    Proceedings of the 39th International Conference on Machine Learning (ICML) , year=

    SkexGen: Autoregressive Generation of CAD Construction Sequences with Disentangled Codebooks , author=. Proceedings of the 39th International Conference on Machine Learning (ICML) , year=

  31. [31]

    Proceedings of the 40th International Conference on Machine Learning , pages=

    Hierarchical neural coding for controllable CAD model generation , author=. Proceedings of the 40th International Conference on Machine Learning , pages=

  32. [32]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    DTGBrepGen: A Novel B-rep Generative Model through Decoupling Topology and Geometry , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  33. [33]

    arXiv preprint arXiv:2405.08609 , year=

    Dynamic nerf: A review , author=. arXiv preprint arXiv:2405.08609 , year=

    work page arXiv

  34. [34]

    2024 IEEE International Symposium on Mixed and Augmented Reality (ISMAR) , pages=

    Is 3dgs useful?: Comparing the effectiveness of recent reconstruction methods in vr , author=. 2024 IEEE International Symposium on Mixed and Augmented Reality (ISMAR) , pages=. 2024 , organization=

    2024

  35. [35]

    Proceedings of the seventh ACM symposium on Solid modeling and applications , pages=

    CSG-BRep duality and compression , author=. Proceedings of the seventh ACM symposium on Solid modeling and applications , pages=

  36. [36]

    Dental materials journal , volume=

    A review of dental CAD/CAM: current status and future perspectives from 20 years of experience , author=. Dental materials journal , volume=. 2009 , publisher=

    2009

  37. [37]

    IEEE Computer Graphics and Applications , volume=

    Ten CAD challenges , author=. IEEE Computer Graphics and Applications , volume=. 2005 , publisher=

    2005

  38. [38]

    Human--Computer Interaction , volume=

    Designing as construction of representations: A dynamic viewpoint in cognitive design research , author=. Human--Computer Interaction , volume=. 2006 , publisher=

    2006

  39. [39]

    Transactions on Graphics , volume=

    Boolean Operation for CAD Models Using a Hybrid Representation , author=. Transactions on Graphics , volume=

    work page

  40. [40]

    CVGIP: image understanding , volume=

    Extracting geometric primitives , author=. CVGIP: image understanding , volume=. 1993 , publisher=

    work page 1993

  41. [41]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=

    Geometric primitives in LiDAR point clouds: A review , author=. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=. 2020 , publisher=

    2020

  42. [42]

    Nature Machine Intelligence , volume=

    Geometric deep learning on molecular representations , author=. Nature Machine Intelligence , volume=. 2021 , publisher=

    2021

  43. [43]

    IEEE Signal Processing Magazine , volume=

    Geometric deep learning: going beyond euclidean data , author=. IEEE Signal Processing Magazine , volume=. 2017 , publisher=

    2017

  44. [44]

    IEEE Access , volume=

    A comprehensive survey on geometric deep learning , author=. IEEE Access , volume=. 2020 , publisher=

    2020

  45. [45]

    ACM SIGGRAPH 2023 conference proceedings , pages=

    Surface and edge detection for primitive fitting of point clouds , author=. ACM SIGGRAPH 2023 conference proceedings , pages=

    2023

  46. [46]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page

  47. [47]

    arXiv preprint arXiv:2105.10620 , year=

    HPNet: Deep Primitive Segmentation Using Hybrid Representations , author=. arXiv preprint arXiv:2105.10620 , year=

    work page arXiv

  48. [48]

    The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Koch, Sebastian and Matveev, Albert and Jiang, Zhongshi and Williams, Francis and Artemov, Alexey and Burnaev, Evgeny and Alexa, Marc and Zorin, Denis and Panozzo, Daniele , title =. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  49. [49]

    arXiv preprint arXiv:2409.16294 , year=

    Gencad: Image-conditioned computer-aided design generation with transformer-based contrastive representation and diffusion priors , author=. arXiv preprint arXiv:2409.16294 , year=

    work page arXiv

  50. [50]

    CAD-MLLM: Unifying multimodality- conditioned CAD generation with MLLM.arXiv preprint arXiv:2411.04954, 2024

    Cad-mllm: Unifying multimodality-conditioned cad generation with mllm , author=. arXiv preprint arXiv:2411.04954 , year=

    work page arXiv

  51. [51]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Cadcrafter: Generating computer-aided design models from unconstrained images , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    work page