arxiv: 2605.11977 · v1 · submitted 2026-05-12 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Optimizing 4D Wires for Sparse 3D Abstraction

Dong-Yi Wu , Tong-Yee Lee

Authors on Pith no claims yet

Pith reviewed 2026-05-13 07:30 UTC · model grok-4.3

classification 💻 cs.CV

keywords 4D wireB-splinegeometric abstractiondifferentiable rendering3D shape representationtopological coherencesparse 3Dwire art

0 comments

The pith

A single continuous 4D B-spline wire captures complex volumetric shapes while enforcing global topological coherence.

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

The paper establishes that one continuous spline curve in four dimensions, carrying both position and a variable width, can represent intricate 3D forms without relying on collections of disconnected segments. This single-curve choice converts local density filling into a global routing task, supplying an inductive bias that favors connected, aesthetically coherent results. A supporting differentiable renderer rasterizes the variable-width curves so that gradient signals from CLIP or score distillation can drive the optimization directly. Demonstrations cover image-conditioned abstraction, multi-view wire generation, and stylized surface coverage, all produced under the same unified representation.

Core claim

A single continuous spline parameterized as a B-spline with spatial coordinates and variable width (x, y, z, w) is sufficiently expressive to capture complex volumetric forms while enforcing global topological coherence; the continuity constraint turns 3D sketching into a global routing problem that yields cleaner aesthetics and improved structural coherence, enabled by a differentiable rendering pipeline that rasterizes variable-width curves with bounded projection error.

What carries the argument

The 4D wire, a single B-spline carrying position (x,y,z) and width (w), together with the differentiable rendering pipeline that projects variable-width curves onto images while keeping projection error bounded.

If this is right

Image-to-3D abstraction yields structures with higher semantic fidelity than collections of independent curves.
Multi-view wire art generation produces outputs with measurably improved structural coherence.
Differentiable stylized surface filling becomes possible under one continuous representation.

Where Pith is reading between the lines

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

Single-curve outputs may be fabricated directly as physical wire sculptures without additional joining steps.
The global-routing bias could extend to other sparse path-planning problems such as network layout or stroke-based rendering.
Variable-width encoding might be generalized to carry additional per-point attributes such as color or material.

Load-bearing premise

The differentiable rendering pipeline for variable-width curves must keep projection error bounded enough to produce stable gradients across the optimization tasks.

What would settle it

Optimize the 4D spline on a standard image-to-3D task; the claim fails if the resulting structure fragments into multiple disconnected pieces or loses semantic match to the input.

Figures

Figures reproduced from arXiv: 2605.11977 by Dong-Yi Wu, Tong-Yee Lee.

**Figure 1.** Figure 1: We compare our single 4D Wire representation against baseline approaches based on independent Bézier [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

**Figure 2.** Figure 2: We compare the reconstructed 3D strokes generated from a single input image (left). Utilizing an equivalent [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: We measure the screen-space error between the projected reference curve, obtained via dense sampling with [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: The 4D Wire as a Semantic Bridge. Stroke width [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Effect of various degrees of freedom. from left to right with increasing keypoints. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 6.** Figure 6: Ablation of key components. Removing either the width-guided reinitialization or the wire refinement module [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗

**Figure 7.** Figure 7: Our 4D wire is less effective than fixed-thickness models at capturing objects with perfectly constant thickness. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Given multi-view prompts (middle), we compare Dreamwire’s discrete curves (left, random colors for each [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 9.** Figure 9: Connection visualization. • Connectivity Cost: To quantify the magnitude of fragmentation in the baseline (DreamWire), we calculate the Minimum Spanning Tree (MST) weight required to bridge all discrete components [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗

**Figure 10.** Figure 10: (a) Our 4D Wire is seamlessly translated into a mesh-based format for advanced rendering. (b) A Shadow [PITH_FULL_IMAGE:figures/full_fig_p015_10.png] view at source ↗

**Figure 11.** Figure 11: Ablation of key components. Removing either the width-guided reinitialization or the wire refinement [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗

**Figure 12.** Figure 12: Failure case in multi-view wire art. Inconsistent prompts across views lead to conflicting geometric [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗

**Figure 13.** Figure 13: Stylized Surface Filling. Top: Input mesh and initialization Noma et al. [2024]. Left: The target style [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗

**Figure 14.** Figure 14: Configurable Surface Filling. The input 3D mesh is initialized using the geodesic curve generation of Noma [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗

read the original abstract

We present a unified framework for 3D geometric abstraction using a single continuous 4D wire, parameterized as a B-spline with spatial coordinates and variable width $(x,y,z,w)$. Existing approaches typically represent shapes as collections of many independent curve segments, which often leads to fragmented structures and limited physical realizability. In contrast, we show that a single continuous spline is sufficiently expressive to capture complex volumetric forms while enforcing global topological coherence. By imposing continuity, our method transforms 3D sketching from a local density-accumulation process into a global routing problem, providing a strong inductive bias toward cleaner aesthetics and improved structural coherence. To enable gradient-based optimization, we introduce a differentiable rendering pipeline that efficiently rasterizes variable-width curves with bounded projection error. This formulation supports robust optimization using modern guidance signals such as Score Distillation Sampling (SDS) or CLIP. We demonstrate applications including image-to-3D abstraction, multi-view wire art generation, and differentiable stylized surface filling. Experiments show that our unified representation produces structures with higher semantic fidelity and improved structural coherence compared to approaches based on collections of discrete curves.

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.

Desk Editor's Note private letter to a colleague

The single 4D B-spline wire is a genuine shift from fragmented curves to one continuous structure, but the rendering pipeline's error behavior is asserted rather than shown.

read the letter

The paper's real move is collapsing many separate curve segments into one 4D B-spline that carries position and width together. That turns the usual local density game into a global routing task, and the continuity constraint does seem to push outputs toward connected, realizable shapes instead of scattered pieces. The applications listed—image-to-3D abstraction, multi-view wire art, and stylized filling—follow directly from being able to optimize the whole thing end-to-end with SDS or CLIP signals. If the renderer actually stays stable, this representation could be a useful inductive bias for anyone who already works with sparse curve models in graphics or generation pipelines. What is missing is any visible derivation or measurement of the projection error. The abstract claims bounded error for variable-width rasterization, yet the stress-test point about tight curvature or rapid width change is not addressed in the provided text. When radius of curvature drops below half the local width, the offset boundary can fold or alias, and nothing in the abstract shows that this stays harmless across optimization steps. Without controls or bounds on that term, it is hard to know whether the reported gains in semantic fidelity come from the 4D formulation or from careful tuning that might not generalize. The experiments are described only at a high level, so the strength of the comparison to discrete-curve baselines is also unclear. This is worth a referee for readers who care about curve-based 3D abstraction and differentiable rendering. The core idea is distinct enough from prior multi-segment work that a careful review could tighten the error analysis and make the method more usable. I would bring it to a reading group to discuss the rendering step, but I would not cite it yet without seeing the full validation.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes representing 3D shapes via a single continuous 4D B-spline wire with coordinates and variable width (x,y,z,w). It claims this unified representation is sufficiently expressive for complex volumetric forms while enforcing global topological coherence, unlike collections of independent curve segments. A differentiable rendering pipeline is introduced to rasterize variable-width curves with bounded projection error, supporting gradient-based optimization with signals such as SDS or CLIP. Demonstrated applications include image-to-3D abstraction, multi-view wire art generation, and differentiable stylized surface filling, with experiments reporting higher semantic fidelity and structural coherence.

Significance. If the bounded-error rendering claim and expressiveness results hold, the work would provide a topologically coherent alternative to fragmented curve collections, with benefits for generative 3D modeling, physical realizability, and stable optimization in graphics pipelines. The reliance on standard B-spline continuity plus external guidance signals is a practical strength that could influence sparse abstraction methods.

major comments (2)

[Abstract and §3] Abstract and §3 (Differentiable Rendering Pipeline): the central claim that the pipeline produces 'bounded projection error' sufficient for stable gradients is load-bearing for the optimization and expressiveness arguments, yet no derivation, explicit error bound, or analysis is supplied for the regime where local curvature radius approaches or falls below w(t)/2 or when dw/dt is large. This leaves the skeptic concern about discontinuities or aliasing in offset curves unaddressed.
[§5] §5 (Experiments): the reported improvements in semantic fidelity and structural coherence over discrete-curve baselines do not include ablations that isolate the single-spline constraint from other factors such as total parameter count or guidance-signal strength, weakening the causal link to the 4D-wire representation.

minor comments (2)

[Introduction] The introduction would benefit from an explicit equation defining the 4D B-spline parameterization (control points, knot vector, degree) to ground the subsequent claims.
[Figures] Figure captions for the rendering pipeline should include a brief note on the approximation method used for thick-curve rasterization.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments highlight important areas for strengthening the theoretical grounding of the rendering pipeline and the experimental isolation of the representation's benefits. We address each point below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Abstract and §3] Abstract and §3 (Differentiable Rendering Pipeline): the central claim that the pipeline produces 'bounded projection error' sufficient for stable gradients is load-bearing for the optimization and expressiveness arguments, yet no derivation, explicit error bound, or analysis is supplied for the regime where local curvature radius approaches or falls below w(t)/2 or when dw/dt is large. This leaves the skeptic concern about discontinuities or aliasing in offset curves unaddressed.

Authors: We acknowledge that the manuscript asserts bounded projection error without supplying a formal derivation or analysis for the critical regimes where curvature radius approaches w(t)/2 or dw/dt becomes large. In the revised version we will add a dedicated error-analysis subsection to §3. This will derive explicit projection-error bounds under these conditions, characterize the continuity of the offset curves, and include both theoretical limits and numerical checks confirming gradient stability. The revision will directly address potential discontinuities and aliasing. revision: yes
Referee: [§5] §5 (Experiments): the reported improvements in semantic fidelity and structural coherence over discrete-curve baselines do not include ablations that isolate the single-spline constraint from other factors such as total parameter count or guidance-signal strength, weakening the causal link to the 4D-wire representation.

Authors: We agree that stronger isolation of the single-spline constraint is needed. The current experiments compare against discrete baselines but do not control for parameter count or guidance-signal strength. In the revision we will add two targeted ablations: (1) a parameter-matched comparison in which the discrete-curve baseline is given an equivalent total number of degrees of freedom, and (2) a controlled sweep of guidance-signal strength (SDS and CLIP) while holding the representation fixed. These results will be reported in §5 to clarify the contribution of topological coherence. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation rests on standard B-spline properties and external guidance

full rationale

The paper's core argument—that a single 4D B-spline is expressive for volumetric forms—follows directly from the continuity and differentiability properties of B-splines, which are standard and externally verifiable. The differentiable rendering pipeline is introduced as a new technical component supporting gradient-based optimization with SDS/CLIP signals; no equations or predictions are shown to reduce by construction to fitted parameters defined on the same data, nor does any load-bearing step rely on self-citation chains or imported uniqueness theorems. The abstract and described formulation remain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The method rests on standard properties of B-splines and introduces a new 4D parameterization whose width channel is treated as an optimizable degree of freedom.

free parameters (1)

B-spline control points and widths
The positions and per-point widths are optimized via gradient descent; no specific fitted constants are stated in the abstract.

axioms (1)

standard math B-splines provide C2 continuity and local support
Invoked to guarantee global topological coherence from a single curve.

invented entities (1)

4D wire (x,y,z,w) no independent evidence
purpose: Encodes both 3D position and local width in one continuous spline
New representation introduced to replace collections of independent curves

pith-pipeline@v0.9.0 · 5489 in / 1297 out tokens · 35269 ms · 2026-05-13T07:30:07.060466+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

a single continuous spline is sufficiently expressive to capture complex volumetric forms while enforcing global topological coherence... transforms 3D sketching from a local density-accumulation process into a global routing problem
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat inductive structure and embed_add echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

uniform cubic B-splines... guarantee C² continuity across the entire wire... knot insertion... preserves the global B-spline formulation

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.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

[1]

2025 , issue_date =

Berio, Daniel and Stroh, Michael and Calinon, Sylvain and Fol Leymarie, Frederic and Deussen, Oliver and Shamir, Ariel , title =. 2025 , issue_date =. doi:10.1145/3763345 , journal =

work page doi:10.1145/3763345 2025
[2]

ACM Trans

Differentiable Vector Graphics Rasterization for Editing and Learning , author =. ACM Trans. Graph. (Proc. SIGGRAPH Asia) , volume =

work page
[3]

arXiv preprint arXiv:2405.15305 , year=

Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering , author =. arXiv preprint arXiv:2405.15305 , year=

work page arXiv
[4]

CVPR , year=

Wired Perspectives: Multi-View Wire Art Embraces Generative AI , author=. CVPR , year=

work page
[5]

2016 , issue_date =

Miguel, Eder and Lepoutre, Mathias and Bickel, Bernd , title =. 2016 , issue_date =. doi:10.1145/2897824.2925978 , journal =

work page doi:10.1145/2897824.2925978 2016
[6]

2022 , issue_date =

Yeh, Chih-Kuo and Le, Thi-Ngoc-Hanh and Hou, Zhi-Ying and Lee, Tong-Yee , title =. 2022 , issue_date =. doi:10.1145/3475798 , journal =

work page doi:10.1145/3475798 2022
[7]

, title =

Liu, Lingjie and Ceylan, Duygu and Lin, Cheng and Wang, Wenping and Mitra, Niloy J. , title =. 2017 , issue_date =. doi:10.1145/3072959.3073682 , journal =

work page doi:10.1145/3072959.3073682 2017
[8]

2020 , issue_date =

Wang, Peng and Liu, Lingjie and Chen, Nenglun and Chu, Hung-Kuo and Theobalt, Christian and Wang, Wenping , title =. 2020 , issue_date =. doi:10.1145/3386569.3392476 , journal =

work page doi:10.1145/3386569.3392476 2020
[9]

2015 , issue_date =

Iarussi, Emmanuel and Li, Wilmot and Bousseau, Adrien , title =. 2015 , issue_date =. doi:10.1145/2816795.2818118 , month =

work page doi:10.1145/2816795.2818118 2015
[10]

2018 , issue_date =

Lira, Wallace and Fu, Chi-Wing and Zhang, Hao , title =. 2018 , issue_date =. doi:10.1145/3272127.3275049 , journal =

work page doi:10.1145/3272127.3275049 2018
[11]

2024 , issue_date =

Choi, Changwoon and Lee, Jaeah and Park, Jaesik and Kim, Young Min , title =. 2024 , issue_date =. doi:10.1145/3658156 , journal =

work page doi:10.1145/3658156 2024
[12]

The conversion matrix between uniform B-spline and Bézier representations , journal =

L Romani and M.A Sabin , keywords =. The conversion matrix between uniform B-spline and Bézier representations , journal =. 2004 , issn =. doi:https://doi.org/10.1016/j.cagd.2004.04.002 , url =

work page doi:10.1016/j.cagd.2004.04.002 2004
[13]

Inserting new knots into B-spline curves , journal =

Wolfgang Boehm , abstract =. Inserting new knots into B-spline curves , journal =. 1980 , issn =. doi:https://doi.org/10.1016/0010-4485(80)90154-2 , url =

work page doi:10.1016/0010-4485(80)90154-2 1980
[14]

Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

Liu, Ruoshi and Wu, Rundi and Van Hoorick, Basile and Tokmakov, Pavel and Zakharov, Sergey and Vondrick, Carl , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

work page 2023
[15]

2018 , issue_date =

Hsiao, Kai-Wen and Huang, Jia-Bin and Chu, Hung-Kuo , title =. 2018 , issue_date =. doi:10.1145/3272127.3275070 , journal =

work page doi:10.1145/3272127.3275070 2018
[16]

2023 , howpublished=

Stable Zero123: Quality 3D Object Generation from Single Images , author=. 2023 , howpublished=

work page 2023
[17]

Mathematische Annalen , year =

Hilbert, David , title =. Mathematische Annalen , year =. doi:10.1007/BF01199431 , url =

work page doi:10.1007/bf01199431
[18]

, title =

Peano, G. , title =. Mathematische Annalen , year =. doi:10.1007/BF01199438 , url =

work page doi:10.1007/bf01199438
[19]

Proceedings of the 4th International Symposium on Non-Photorealistic Animation and Rendering , pages =

Pedersen, Hans and Singh, Karan , title =. Proceedings of the 4th International Symposium on Non-Photorealistic Animation and Rendering , pages =. 2006 , isbn =. doi:10.1145/1124728.1124742 , abstract =

work page doi:10.1145/1124728.1124742 2006
[20]

2024 , isbn =

Tojo, Kenji and Shamir, Ariel and Bickel, Bernd and Umetani, Nobuyuki , title =. 2024 , isbn =. doi:10.1145/3641519.3657453 , booktitle =

work page doi:10.1145/3641519.3657453 2024
[21]

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

Wang, Zhenyu and Lu, Min , title =. Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , articleno =. 2025 , isbn =. doi:10.1145/3721238.3730690 , abstract =

work page doi:10.1145/3721238.3730690 2025
[22]

Surface-Filling Curve Flows via Implicit Medial Axes , year =

Noma, Yuta and Sell\'. Surface-Filling Curve Flows via Implicit Medial Axes , year =. ACM Trans. Graph. , month = jul, articleno =. doi:10.1145/3658158 , abstract =

work page doi:10.1145/3658158
[23]

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

CLIPAG: Towards Generator-Free Text-to-Image Generation , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

work page
[24]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

Kwon, Gihyun and Ye, Jong Chul , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

work page 2022
[25]

Advances in Neural Information Processing Systems (NeurIPS) , volume=

CLIPDraw: Exploring Text-to-Drawing Synthesis through Language-Image Encoders , author=. Advances in Neural Information Processing Systems (NeurIPS) , volume=

work page
[26]

ACM Transactions on Graphics (TOG) , volume=

CLIPasso: Semantically-Aware Object Sketching , author=. ACM Transactions on Graphics (TOG) , volume=. 2022 , publisher=

work page 2022
[27]

International Conference on Machine Learning , pages=

Learning Transferable Visual Models From Natural Language Supervision , author=. International Conference on Machine Learning , pages=. 2021 , organization=

work page 2021
[28]

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

VectorFusion: Text-to-SVG by Abstracting Pixel-Based Diffusion Models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

work page
[29]

Advances in Neural Information Processing Systems (NeurIPS) , volume=

DiffSketcher: Text Guided Vector Sketch Synthesis through Latent Diffusion Models , author=. Advances in Neural Information Processing Systems (NeurIPS) , volume=

work page
[30]

2018 , issue_date =

Zhao, Haisen and Zhang, Hao and Xin, Shiqing and Deng, Yuanmin and Tu, Changhe and Wang, Wenping and Cohen-Or, Daniel and Chen, Baoquan , title =. ACM Trans. Graph. , month = jul, articleno =. 2018 , issue_date =. doi:10.1145/3197517.3201338 , abstract =

work page doi:10.1145/3197517.3201338 2018
[31]

The Twelfth International Conference on Learning Representations , year=

DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D Generation , author=. The Twelfth International Conference on Learning Representations , year=

work page
[32]

Proceedings of the 38th International Conference on Machine Learning , pages=

Learning Transferable Visual Models From Natural Language Supervision , author=. Proceedings of the 38th International Conference on Machine Learning , pages=. 2021 , organization=

work page 2021
[33]

International Conference on Learning Representations , year=

A Neural Representation of Sketch Drawings , author=. International Conference on Learning Representations , year=

work page
[34]

and Shechtman, Eli and Wang, Oliver , title =

Zhang, Richard and Isola, Phillip and Efros, Alexei A. and Shechtman, Eli and Wang, Oliver , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , month =

work page
[35]

Curves and Surfaces for CAGD (Fifth Edition) , publisher =

4 - The de Casteljau Algorithm , editor =. Curves and Surfaces for CAGD (Fifth Edition) , publisher =. 2002 , series =. doi:https://doi.org/10.1016/B978-155860737-8/50004-1 , url =

work page doi:10.1016/b978-155860737-8/50004-1 2002
[36]

ACM Trans

Iluz, Shir and Vinker, Yael and Hertz, Amir and Berio, Daniel and Cohen-Or, Daniel and Shamir, Ariel , title =. ACM Trans. Graph. , month = jul, articleno =. 2023 , issue_date =. doi:10.1145/3592123 , abstract =

work page doi:10.1145/3592123 2023
[37]

The Eleventh International Conference on Learning Representations , year=

DreamFusion: Text-to-3D using 2D Diffusion , author=. The Eleventh International Conference on Learning Representations , year=

work page
[38]

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

Towards layer-wise image vectorization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[39]

2025 , isbn =

Faruqi, Faraz and Paonaskar, Josha and Schuler, Riley and Prevey, Aiden and Taylor, Carson and Tak, Anika and Guinto, Anthony and Shilamkar, Eeshani and Cheenaruenthong, Natarith and Nisser, Martin , title =. 2025 , isbn =. doi:10.1145/3745778.3766662 , booktitle =

work page doi:10.1145/3745778.3766662 2025
[40]

, title =

Laurentini, A. , title =. 1994 , issue_date =. doi:10.1109/34.273735 , journal =

work page doi:10.1109/34.273735 1994
[41]

2022 , howpublished=

Stable Diffusion v1.5 Model Card , author=. 2022 , howpublished=

work page 2022
[42]

1999 , publisher=

Vision Science: Photons to Phenomenology , author=. 1999 , publisher=

work page 1999