Pith. sign in

REVIEW 3 major objections 6 minor 54 references

ViewCraft3D: High-Fidelity and View-Consistent 3D Vector Graphics Synthesis

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read 3D vector graphics can be made view-consistent by fitting Bézier curves directly in 3D space, and the paper reports this takes about 30 minutes.

desk verdict A plausible new pipeline for 3D vector graphics from single images, but the evaluation and the TripoSG VAE assumption need serious work before the headline claims are credible. read the letter →

arxiv 2505.19492 v1 pith:HJDVNJX3 submitted 2025-05-26 cs.CV

Chuang Wang , Haitao Zhou , Ling Luo , Qian Yu This is my paper · ORCID
classification cs.CV
keywords 3DvectorgraphicsBéziercurvefittingviewconsistencyscoredistillationsamplingimage-to-3Dreconstructionsingle-imagegenerationpointcloudclusteringsketchabstraction
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a sparse set of 3D Bézier curves, fitted to a reconstructed mesh and then refined with a 3D diffusion prior, can capture an object's essential structure from a single image while staying consistent across viewpoints. Prior methods optimize curves through 2D projections, so the same stroke disagrees between views and takes two to six hours; VC3D optimizes the curve control points directly in 3D space, which makes view consistency a property of the representation rather than something enforced per view. The paper reports that the resulting sketches, typically fewer than 100 curves, beat the two baselines on semantic and aesthetic scores and complete in about 30 minutes. If correct, this makes 3D vector graphics practical for VR, shape retrieval, and conceptual design from ordinary photos.

What carries the argument

The load-bearing object is the 3D Bézier curve set $\{C_i\}_{i=1}^n$, with control points in $\mathbb{R}^3$, treated as a differentiable point cloud through uniform sampling of $B(t)$. The mechanism is a two-stage procedure: geometric fitting by a Chamfer Distance loss between sampled curve points and the salient point cloud, then score distillation refinement in the latent space of a pretrained image-to-3D model. A perspective projection identity — a 3D Bézier curve projects to a 2D rational Bézier curve — lets the same 3D curves be rendered into any viewpoint for evaluation, while the optimization itself stays in 3D space.

What would settle it

Render the final curves from the 12 evaluation viewpoints and also from held-out viewpoints; if strokes that should be a single 3D line appear misaligned or duplicate across adjacent views as badly as the 2D-projection baselines, the consistency-by-construction claim is contradicted. More directly, feed the sparse combined point cloud of Stage II to the pretrained VAE and compare its latent to the latent of the dense reconstructed mesh: if the sparse latent carries no recoverable structure, the SDS refinement is not doing what the paper says.

Watch

Extended reading notes

Core claim

The central claim is that view-consistent 3D vector graphics can be produced by a two-stage pipeline that never leaves 3D space. In Stage I, an image-to-3D model reconstructs a mesh, a sharp-edge-sampling process extracts a salient point cloud, points are clustered by spatial proximity and orientation, and each cluster is fit with a cubic Bézier curve under a Chamfer Distance loss. In Stage II, the first curves are frozen, new curves are initialized only in regions the point cloud covered poorly, and a score distillation loss uses the same image-to-3D model's latent space to refine the new parameters. The paper reports that this two-stage method outperforms the 2D-projection-based baselines on CLIPScore and aesthetic score, reduces generation time to about 0.5 hours versus 2 to 6 hours, and keeps the same 3D elements consistent from every viewpoint because the optimization happens once in 3D rather than separately per view.

Load-bearing premise

The refinement stage assumes that the latent space of the pretrained image-to-3D model gives useful feedback about a sparse point cloud sampled from fewer than 100 Bézier curves, even though that model was trained on dense 3D shapes.

Editorial extensions

If this is right

  • A single photo can produce a view-consistent 3D sketch in about 30 minutes, making interactive and iterative sketching feasible on one GPU.
  • Because consistency is structural, the generated curves need no per-view post-processing; rendering from arbitrary new viewpoints is safe.
  • By raising the cluster filtering threshold, the same pipeline produces coarser, more abstract sketches, so abstraction level is a controllable parameter.
  • The same mesh-to-curve fitting could be applied at scale to existing mesh datasets, building large 3D vector graphics collections for training or retrieval.
  • The reported gains on CLIPScore and aesthetic score, together with the ablation, identify both geometric fitting and SDS refinement as contributing to the final quality.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next step the paper leaves open is ordering curves by depth and giving them per-stroke opacity; the reconstructed mesh already provides the depth information needed for occlusion-aware rendering.
  • Since the pipeline inherits the reconstructed mesh, sketch quality is bounded by mesh quality; comparing VC3D with fitting on ground-truth meshes would separate reconstruction errors from fitting errors.
  • The Stage I fitting is nearly deterministic, so most of the 30-minute cost sits in the SDS refinement; a feed-forward predictor of curve control points could plausibly cut the time further.
  • The reported superiority over the 2D-projection baselines may be partly due to different initialization and supervision; an ablation that gives the baselines the same reconstructed mesh would isolate the contribution of optimizing in 3D.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. ViewCraft3D (VC3D) is a two-stage method for generating 3D vector graphics, represented as 3D Bézier curves, from a single input image. Stage I reconstructs a 3D mesh with a pretrained image-to-3D model (TripoSG), extracts a salient point cloud via sharp-edge and silhouette sampling, clusters points using spatial proximity and local-PCA orientation, and fits cubic Bézier curves by optimizing a Chamfer distance. Stage II freezes these curves, adds new curves in under-covered regions, samples a combined point cloud, encodes it with TripoSG's VAE, and refines only the new curve parameters using SDS loss conditioned on the input image. The paper reports CLIPScore and Aesthetic Score on 40 images, an ablation of the two stages, and a runtime of about 0.5 hours, compared with about 2 hours for Diff3DS and about 6 hours for 3Doodle.

Significance. The central idea is attractive: optimizing curve parameters directly in 3D, rather than through per-view 2D projections, should make view consistency structural rather than emergent. The geometric fitting stage is a sensible way to convert a mesh into a sparse curve abstraction, and the reported runtime advantage is a concrete strength of the work. However, the evidence is not yet commensurate with the paper's claims: the refinement stage rests on an unspecified and likely unsupported use of a pretrained 3D VAE on sparse point clouds; the quantitative evaluation covers few baselines, no statistical variability, and no view-consistency metric; and the region-selection algorithm for Stage II is not described. The method's potential is clear, but the manuscript needs substantial additional evidence and specification.

major comments (3)
  1. [Section 3.3, Eq. (4)] The refinement stage assumes that the pretrained TripoSG VAE can encode a sparse point cloud Pcombined = Pc ∪ Pc′ sampled from fewer than 100 Bézier curves and that the resulting latent z supports meaningful denoising. The manuscript never states what input representation TripoSG's VAE expects or how Pcombined is converted into it. If the encoder expects a dense volumetric field (e.g., an SDF grid or triplane), then feeding a point cloud is out of distribution and the SDS gradients in Eq. (4) would not reflect a valid 3D prior. Please specify the input conversion and provide validation, for example by measuring reconstruction error or denoising quality at the relevant point densities.
  2. [Section 4.2, Table 1] The claim that VC3D outperforms previous methods is not fully supported because Dream3DVG [47], which is discussed in Related Work, is not included in the evaluation. The metrics are also reported as point averages over 40 images without error bars or significance tests, and no view-consistency metric is provided although view consistency is a headline contribution. Please add the missing baseline, report per-image statistics, and include a quantitative view-consistency measure across the 12 rendered views.
  3. [Section 3.3] The initialization of Stage II is not described precisely enough to reproduce the method. The text only says that under-covered regions are identified by 'analyzing the mesh's vertex distribution,' and it does not specify how the new curves are initialized, how many new curves n′ are added, or what makes a region 'intricate to approximate.' Because the improvement attributed to this stage in Table 2 depends entirely on these choices, the paper should state the selection and initialization algorithm explicitly.
minor comments (6)
  1. [Section 3.2.2] The value of k for the k-nearest-neighbor PCA orientation estimation is not reported in the implementation details.
  2. [Section 3.2.3, Eq. (3)] The balance weight λ in the Chamfer distance loss is not given in Section 4.1.
  3. [Section 4.1] The number of new curves n′ in Stage II and the criterion for selecting their initial locations are not specified; please add these along with the values of the other hyperparameters.
  4. [Section 4.2] The '12 views using identical camera parameters' are not defined (e.g., azimuth/elevation range), so the view-consistency evaluation is hard to interpret.
  5. [Section 4.3, Table 2] Variant 1 is described as having only point extraction and clustering; it is unclear what is rendered and scored to produce the reported CLIPScore and Aesthetic Score.
  6. [Section 4.2.1, Figure 5] The caption notes that 3Doodle uses 120 rendered images of the reconstructed mesh while Diff3DS and VC3D use the single input image; the authors should discuss whether this difference in input modality affects the comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: VC3D is a fitting/generation pipeline whose output is optimized to match the input by construction, and the efficiency claim is independent of any self-citation.

full rationale

VC3D is a generative fitting pipeline, not a derivation with independent predictive claims. Stage I reconstructs a mesh from the input image and fits 3D Bézier curves to salient points of that mesh via Chamfer distance; matching the input is the optimization objective, so fidelity to the source image is by construction and is not presented as an independent prediction. Stage II refines newly initialized curves using SDS with TripoSG's VAE encoder and denoising model conditioned on the same image. This is a design choice and raises an out-of-distribution concern (the VAE input representation for a sparse point cloud is unspecified), but it is not circular: the SDS gradients come from a pretrained model, not from the paper's own outputs, and no equation reduces Stage II's loss to Stage I's fitting loss. View consistency follows from optimizing curves directly in 3D and projecting them; this is a representational property, not a self-referential claim. The paper's self-citations (DiffSketcher, SVGDreamer, SVGFusion, VectorPainter, etc.) appear only as related work and are not load-bearing for the method's correctness or for any asserted theorem. Quantitative metrics use external CLIP and aesthetic predictors on rendered views, not parameters fitted by the paper. I find no step where a claimed prediction or first-principles result is equivalent to its input by definition.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method assembles existing components (TripoSG, Sharp Edge Sampling, EdgeGaussians clustering, DiffVG, SDS) into a new pipeline. The ledger is dominated by hand-set clustering thresholds and an unverified assumption about encoding sparse point clouds with a shape VAE.

free parameters (6)
  • dthresh = 0.05
    Distance threshold for point clustering in Stage I (Section 4.1). Controls which points are grouped into curve clusters.
  • theta_thresh = 50 degrees
    Angle threshold for orientation compatibility in clustering (Section 4.1). Determines whether edges align into a cluster.
  • tau = 10 (default); varied 10-40 in ablation
    Filter size for small clusters in Section 3.2.2; directly controls the number of Bézier curves (Figure 8).
  • lambda_chamfer
    Weight balancing the two Chamfer distance terms in Eq. (3); value not reported in the paper.
  • SDS_weight = 2e-4
    Weight for SDS loss in Stage II, set in Section 4.1.
  • k_nearest_neighbors
    Number of neighbors for PCA orientation estimation in Section 3.2.2; not specified.
assumptions (4)
  • domain assumption TripoSG's reconstructed 3D mesh faithfully represents the input object's geometric structure, including sharp edges and silhouettes.
    The entire approach derives curves from this mesh; if reconstruction is poor, the vector graphic inherits errors. Invoked in Section 3.1.
  • domain assumption The VAE encoder of TripoSG can meaningfully encode a sparse point cloud sampled from Bézier curves into a latent space in which SDS gradients improve curve placement.
    Stage II relies on this to refine new curves; no validation is provided. Invoked in Section 3.3.
  • domain assumption SDS gradients from the image-conditioned 3D diffusion model are an effective supervisor for geometric fidelity of a point cloud.
    Assumes the diffusion prior can judge curve collections, not just dense shapes. Invoked in Eq. (4).
  • domain assumption CLIPScore and Aesthetic Score are valid proxies for 'high fidelity' and 'view consistency'.
    Used as the only quantitative evidence in Section 4.2.2; no view-consistency metric is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ViewCraft3D: High-Fidelity and View-Consistent 3D Vector Graphics Synthesis." pith.science (2026). https://pith.science/paper/HJDVNJX3

@misc{pith2026250519492,
  author       = {Pith},
  title        = {Pith review of: ViewCraft3D: High-Fidelity and View-Consistent 3D Vector Graphics Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HJDVNJX3}},
  note         = {Machine review of arXiv:2505.19492}
}
read the original abstract

3D vector graphics play a crucial role in various applications including 3D shape retrieval, conceptual design, and virtual reality interactions due to their ability to capture essential structural information with minimal representation. While recent approaches have shown promise in generating 3D vector graphics, they often suffer from lengthy processing times and struggle to maintain view consistency. To address these limitations, we propose ViewCraft3D (VC3D), an efficient method that leverages 3D priors to generate 3D vector graphics. Specifically, our approach begins with 3D object analysis, employs a geometric extraction algorithm to fit 3D vector graphics to the underlying structure, and applies view-consistent refinement process to enhance visual quality. Our comprehensive experiments demonstrate that VC3D outperforms previous methods in both qualitative and quantitative evaluations, while significantly reducing computational overhead. The resulting 3D sketches maintain view consistency and effectively capture the essential characteristics of the original objects.

Figures

Figures reproduced from arXiv: 2505.19492 by the authors.

Figure 1
Figure 1. We propose ViewCraft3D (VC3D), a method to generate 3D vector graphics from a single image. VC3D can leverage 3D prior knowledge to generate high-quality and view-consistent 3D vector graphics. Abstract 3D vector graphics play a crucial role in various applications including 3D shape retrieval, conceptual design, and virtual reality interactions due to their ability to capture essential structural information with m… view at source ↗
Figure 2
Figure 2. Examples of VR sketches [31]. Recent years have witnessed remarkable progress in 2D vector graphics generation. Works like CLI￾Passo [34] and CLIPDraw [7] pioneered the use of CLIP’s visual-semantic understanding to guide vector graphics optimization. Building on these foundations, methods such as VectorFusion [11], DiffSketcher [41], and SVGDreamer [43] further leveraged diffusion models to achieve higher fi￾delity… view at source ↗
Figure 3
Figure 3. The overall architecture of the proposed method, showcasing the initial generation of 3D [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The visualization process of Point Cloud Clustering. Each point is assigned an orientation [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of different methods. Diff3DS and VC3D use a single image [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Illustration of the optimization process using Chamfer Distance loss [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the optimization effect of SDS loss [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Effect of the filtering threshold in Point Cloud Clustering. The number of Bézier curves [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 29 canonical work pages

  1. [47]

    Empowering vector graphics with consistently arbitrary viewing and view-dependent visibility

    Li Yidi, Xiao Jun, Lu Zhengda, Wang Yiqun, and Jiang Haiyong. Empowering vector graphics with consistently arbitrary viewing and view-dependent visibility. pages 1–14, 2025

    work page 2025

  2. [1]

    Mid-air drawing of curves on 3d surfaces in virtual reality

    Rahul Arora and Karan Singh. Mid-air drawing of curves on 3d surfaces in virtual reality. ACM Transactions on Graphics (TOG), 40(3):1–17, 2021

    work page 2021

  3. [2]

    Ilovesketch: as-natural-as-possible sketching system for creating 3d curve models

    Seok-Hyung Bae, Ravin Balakrishnan, and Karan Singh. Ilovesketch: as-natural-as-possible sketching system for creating 3d curve models. In Proceedings of the 21st annual ACM symposium on User interface software and technology, pages 151–160, 2008

    work page 2008

  4. [3]

    Edgegaussians-3d edge mapping via gaussian splatting

    Kunal Chelani, Assia Benbihi, Torsten Sattler, and Fredrik Kahl. Edgegaussians-3d edge mapping via gaussian splatting. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3268–3279. IEEE, 2025

    work page 2025

  5. [4]

    Dora: Sampling and benchmarking for 3d shape variational auto-encoders

    Rui Chen, Jianfeng Zhang, Yixun Liang, Guan Luo, Weiyu Li, Jiarui Liu, Xiu Li, Xiaoxiao Long, Jiashi Feng, and Ping Tan. Dora: Sampling and benchmarking for 3d shape variational auto-encoders. arXiv preprint arXiv:2412.17808, 2024

    arXiv 2024

  6. [5]

    3doodle: Compact abstraction of objects with 3d strokes

    Changwoon Choi, Jaeah Lee, Jaesik Park, and Young Min Kim. 3doodle: Compact abstraction of objects with 3d strokes. ACM Transactions on Graphics (TOG), 43(4):1–13, 2024

    work page 2024

  7. [6]

    Scaling rectified flow trans- formers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow trans- formers for high-resolution image synthesis. In Forty-first international conference on machine learning, 2024

    2024

  8. [7]

    Clipdraw: Exploring text-to-drawing synthe- sis through language-image encoders

    Kevin Frans, Lisa Soros, and Olaf Witkowski. Clipdraw: Exploring text-to-drawing synthe- sis through language-image encoders. Advances in Neural Information Processing Systems, 35:5207–5218, 2022

    work page 2022

Show all 54 references
  1. [8]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  2. [9]

    Vectorpainter: Advanced stylized vector graphics synthesis using stroke-style priors

    Juncheng Hu, Ximing Xing, Jing Zhang, and Qian Yu. Vectorpainter: Advanced stylized vector graphics synthesis using stroke-style priors. arXiv preprint arXiv:2405.02962, 2024

    2024 arXiv

  3. [10]

    Word-as- image for semantic typography

    Shir Iluz, Yael Vinker, Amir Hertz, Daniel Berio, Daniel Cohen-Or, and Ariel Shamir. Word-as- image for semantic typography. ACM Transactions on Graphics (TOG), 42(4):1–11, 2023

    2023

  4. [11]

    Vectorfusion: Text-to-svg by abstracting pixel-based diffusion models

    Ajay Jain, Amber Xie, and Pieter Abbeel. Vectorfusion: Text-to-svg by abstracting pixel-based diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1911–1920, 2023

    1911

  5. [12]

    3d gaussian splatting for real-time radiance field rendering

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

    2023

  6. [13]

    Recovering dynamic 3d sketches from videos

    Jaeah Lee, Changwoon Choi, Young Min Kim, and Jaesik Park. Recovering dynamic 3d sketches from videos. arXiv preprint arXiv:2503.20321, 2025

    2025 arXiv

  7. [14]

    Differentiable vector graphics rasterization for editing and learning

    Tzu-Mao Li, Michal Luká ˇc, Michaël Gharbi, and Jonathan Ragan-Kelley. Differentiable vector graphics rasterization for editing and learning. ACM Transactions on Graphics (TOG), 39(6):1–15, 2020

    2020

  8. [15]

    Triposg: High-fidelity 3d shape synthesis using large-scale rectified flow models

    Yangguang Li, Zi-Xin Zou, Zexiang Liu, Dehu Wang, Yuan Liang, Zhipeng Yu, Xingchao Liu, Yuan-Chen Guo, Ding Liang, Wanli Ouyang, et al. Triposg: High-fidelity 3d shape synthesis using large-scale rectified flow models. arXiv preprint arXiv:2502.06608, 2025

    2025 arXiv

  9. [16]

    One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization

    Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Mukund Varma T, Zexiang Xu, and Hao Su. One-2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization. Advances in Neural Information Processing Systems, 36:22226–22246, 2023

    2023

  10. [17]

    Zero-1-to-3: Zero-shot one image to 3d object

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl V ondrick. Zero-1-to-3: Zero-shot one image to 3d object. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9298–9309, 2023. 10

    2023

  11. [18]

    Wonder3d: Single image to 3d using cross-domain diffusion

    Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, et al. Wonder3d: Single image to 3d using cross-domain diffusion. In Proceedings of the IEEE/CVF conference on computer vision and patter...

    2024

  12. [19]

    3d vr sketch guided 3d shape prototyping and exploration

    Ling Luo, Pinaki Nath Chowdhury, Tao Xiang, Yi-Zhe Song, and Yulia Gryaditskaya. 3d vr sketch guided 3d shape prototyping and exploration. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9267–9276, 2023

    2023

  13. [20]

    Towards 3d vr-sketch to 3d shape retrieval

    Ling Luo, Yulia Gryaditskaya, Yongxin Yang, Tao Xiang, and Yi-Zhe Song. Towards 3d vr-sketch to 3d shape retrieval. In 2020 International Conference on 3D Vision (3DV), pages 81–90. IEEE, 2020

    2020

  14. [21]

    Karl Pearson. Liii. on lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin philosophical magazine and journal of science, 2(11):559–572, 1901

    1901

  15. [22]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    2023

  16. [23]

    Dreamfusion: Text-to-3d using 2d diffusion

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

    2022 arXiv

  17. [24]

    Pointnet++: Deep hierarchical feature learning on point sets in a metric space

    Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. Advances in neural information processing systems, 30, 2017

    2017

  18. [25]

    Wired perspectives: Multi-view wire art embraces generative ai

    Zhiyu Qu, Lan Yang, Honggang Zhang, Tao Xiang, Kaiyue Pang, and Yi-Zhe Song. Wired perspectives: Multi-view wire art embraces generative ai. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6149–6158, 2024

    2024

  19. [26]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pa...

    2021

  20. [27]

    A stochastic approximation method

    Herbert Robbins and Sutton Monro. A stochastic approximation method. The annals of mathematical statistics, pages 400–407, 1951

    1951

  21. [28]

    High- resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High- resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  22. [29]

    Improved aesthetic predictor, 2022

    Christoph Schuhmann. Improved aesthetic predictor, 2022

    2022

  23. [30]

    Mvdream: Multi- view diffusion for 3d generation

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi- view diffusion for 3d generation. arXiv preprint arXiv:2308.16512, 2023

    2023 arXiv

  24. [31]

    Sketchfab - the platform for 3d and ar on the web, 2023

    Sketchfab. Sketchfab - the platform for 3d and ar on the web, 2023. 3D VR sketch

    2023

  25. [32]

    Fabricable 3d wire art

    Kenji Tojo, Ariel Shamir, Bernd Bickel, and Nobuyuki Umetani. Fabricable 3d wire art. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  26. [33]

    Clipascene: Scene sketching with different types and levels of abstraction

    Yael Vinker, Yuval Alaluf, Daniel Cohen-Or, and Ariel Shamir. Clipascene: Scene sketching with different types and levels of abstraction. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4146–4156, 2023

    2023

  27. [34]

    Clipasso: Semantically-aware object sketching

    Yael Vinker, Ehsan Pajouheshgar, Jessica Y Bo, Roman Christian Bachmann, Amit Haim Bermano, Daniel Cohen-Or, Amir Zamir, and Ariel Shamir. Clipasso: Semantically-aware object sketching. ACM Transactions on Graphics (TOG), 41(4):1–11, 2022

    2022

  28. [35]

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

    Kailu Wu, Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan, and Kaisheng Ma. Unique3d: High-quality and efficient 3d mesh generation from a single image. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. 11

    2024

  29. [36]

    Chat2svg: Vector graphics generation with large language models and image diffusion models

    Ronghuan Wu, Wanchao Su, and Jing Liao. Chat2svg: Vector graphics generation with large language models and image diffusion models. arXiv preprint arXiv:2411.16602, 2024

    2024

  30. [37]

    Iconshop: Text-guided vector icon synthesis with autoregressive transformers

    Ronghuan Wu, Wanchao Su, Kede Ma, and Jing Liao. Iconshop: Text-guided vector icon synthesis with autoregressive transformers. ACM Transactions on Graphics (TOG), 42(6):1–14, 2023

    2023

  31. [38]

    Structured 3d latents for scalable and versatile 3d generation

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d generation. arXiv preprint arXiv:2412.01506, 2024

    2024 arXiv

  32. [39]

    Empowering llms to understand and generate complex vector graphics

    Ximing Xing, Juncheng Hu, Guotao Liang, Jing Zhang, Dong Xu, and Qian Yu. Empowering llms to understand and generate complex vector graphics. arXiv preprint arXiv:2412.11102, 2024

    2024 arXiv

  33. [40]

    Svgfusion: Scalable text-to-svg generation via vector space diffusion

    Ximing Xing, Juncheng Hu, Jing Zhang, Dong Xu, and Qian Yu. Svgfusion: Scalable text-to-svg generation via vector space diffusion. arXiv preprint arXiv:2412.10437, 2024

    2024 arXiv

  34. [41]

    Diffsketcher: Text guided vector sketch synthesis through latent diffusion models

    XiMing Xing, Chuang Wang, Haitao Zhou, Jing Zhang, Qian Yu, and Dong Xu. Diffsketcher: Text guided vector sketch synthesis through latent diffusion models. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Information Processin...

    2023

  35. [42]

    Svgdreamer++: Advancing editability and diversity in text-guided svg generation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Ximing Xing, Qian Yu, Chuang Wang, Haitao Zhou, Jing Zhang, and Dong Xu. Svgdreamer++: Advancing editability and diversity in text-guided svg generation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    2025

  36. [43]

    Svgdreamer: Text guided svg generation with diffusion model

    Ximing Xing, Haitao Zhou, Chuang Wang, Jing Zhang, Dong Xu, and Qian Yu. Svgdreamer: Text guided svg generation with diffusion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4546–4555, 2024

    2024

  37. [44]

    Hunyuan3d 1.0: A unified framework for text-to-3d and image-to-3d generation

    Xianghui Yang, Huiwen Shi, Bowen Zhang, Fan Yang, Jiacheng Wang, Hongxu Zhao, Xinhai Liu, Xinzhou Wang, Qingxiang Lin, Jiaao Yu, et al. Hunyuan3d 1.0: A unified framework for text-to-3d and image-to-3d generation. arXiv preprint arXiv:2411.02293, 2024

    2024 arXiv

  38. [45]

    Omnisvg: A unified scalable vector graphics generation model

    Yiying Yang, Wei Cheng, Sijin Chen, Xianfang Zeng, Jiaxu Zhang, Liao Wang, Gang Yu, Xingjun Ma, and Yu-Gang Jiang. Omnisvg: A unified scalable vector graphics generation model. arXiv preprint arXiv:2504.06263, 2025

    2025

  39. [46]

    Hi3dgen: High-fidelity 3d geometry generation from images via normal bridging

    Chongjie Ye, Yushuang Wu, Ziteng Lu, Jiahao Chang, Xiaoyang Guo, Jiaqing Zhou, Hao Zhao, and Xiaoguang Han. Hi3dgen: High-fidelity 3d geometry generation from images via normal bridging. arXiv preprint arXiv:2503.22236, 3, 2025

    2025 arXiv

  40. [48]

    Piecewise- smooth surface fitting onto unstructured 3d sketches

    Emilie Yu, Rahul Arora, J Andreas Baerentzen, Karan Singh, and Adrien Bousseau. Piecewise- smooth surface fitting onto unstructured 3d sketches. ACM Transactions on Graphics (TOG), 41(4):1–16, 2022

    2022

  41. [49]

    Cassie: Curve and surface sketching in immersive environments

    Emilie Yu, Rahul Arora, Tibor Stanko, J Andreas Bærentzen, Karan Singh, and Adrien Bousseau. Cassie: Curve and surface sketching in immersive environments. In Proceedings of the 2021 CHI conference on human factors in computing systems, pages 1–14, 2021

    2021

  42. [50]

    Videodoodles: Hand-drawn animations on videos with scene-aware canvases

    Emilie Yu, Kevin Blackburn-Matzen, Cuong Nguyen, Oliver Wang, Rubaiat Habib Kazi, and Adrien Bousseau. Videodoodles: Hand-drawn animations on videos with scene-aware canvases. ACM Transactions on Graphics (TOG), 42(4):1–12, 2023

    2023

  43. [51]

    3d-layers: Bringing layer- based color editing to vr painting

    Emilie Yu, Fanny Chevalier, Karan Singh, and Adrien Bousseau. 3d-layers: Bringing layer- based color editing to vr painting. ACM Transactions on Graphics (TOG), 43(4):1–15, 2024

    2024

  44. [52]

    Scaffoldsketch: Accurate industrial design drawing in vr

    Xue Yu, Stephen DiVerdi, Akshay Sharma, and Yotam Gingold. Scaffoldsketch: Accurate industrial design drawing in vr. InThe 34th Annual ACM Symposium on User Interface Software and Technology, pages 372–384, 2021. 12

    2021

  45. [53]

    Diff3ds: Generating view-consistent 3d sketch via differentiable curve rendering

    Yibo Zhang, Lihong Wang, Changqing Zou, Tieru Wu, and Rui Ma. Diff3ds: Generating view-consistent 3d sketch via differentiable curve rendering. arXiv preprint arXiv:2405.15305, 2024

    2024 arXiv

  46. [54]

    Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation

    Zibo Zhao, Zeqiang Lai, Qingxiang Lin, Yunfei Zhao, Haolin Liu, Shuhui Yang, Yifei Feng, Mingxin Yang, Sheng Zhang, Xianghui Yang, et al. Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation. arXiv preprint arXiv:2501.12202, 2025. 13

    2025 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.