Pith. sign in

REVIEW 4 major objections 5 minor 37 references

INST-Sculpt: Interactive Stroke-based Neural SDF Sculpting

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Stroke-based sculpting can be applied directly to neural SDFs on the fly

desk verdict Genuinely new stroke-based sculpting for neural SDFs, with real timing wins over 3DNS; evaluation is a bit self-referential and missing cumulative stability, but the core idea holds. read the letter →

arxiv 2502.02891 v1 pith:7WQMZP6R submitted 2025-02-05 cs.GR cs.CV

Fizza Rubab , Yiying Tong This is my paper · ORCID
classification cs.GRcs.CV
keywords neuralsigneddistancefunctionsimplicitrepresentationsstroke-basedsculptingtubularsamplingSIRENinteractiveshapeeditingbrushprofiles3Dmodeling
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

INST-Sculpt claims that neural signed distance functions, whose geometry is hidden inside neural network weights, can be sculpted interactively with stroke-based brushes rather than isolated point edits. The method builds a tubular neighborhood around a user-drawn curve on the surface, offsets sampled points along the surface normal by a customizable brush profile, and fine-tunes the SDF network so its zero level set follows the displaced points. It reports that a single stroke edit takes under a second, achieves lower Chamfer error than the 3DNS point-edit baseline, and runs up to 16 times faster by sampling far fewer points. If correct, this would let artists edit neural implicit shapes with the same stroke-based carving and extruding intuition used in mesh sculpting, without ever converting to another representation.

What carries the argument

The load-bearing mechanism is tubular sampling in a moving u-v-n coordinate frame, combined with a user-defined brush profile. For each point along the stroke, the method samples uniformly spaced offsets in the v-direction within a brush radius, projects them onto the original zero level set, and displaces them along the surface normal by the product m(u) B_{r,s}(v), where B_{r,s} is a normalized brush profile such as a Catmull-Rom spline through user control points and m(u) is a modulation function along the stroke. Fine-tuning the SIREN MLP to put the displaced points p' on the zero level set converts a continuous user stroke into a small, targeted training set, which is what makes the edit both local and fast.

What would settle it

Take a sphere and apply a single stroke using a brush profile with several sharp oscillations, such as five sine peaks within the brush radius, at the default 2x128 SIREN capacity; if the edited zero level set inside the tube merges those peaks and its Chamfer distance in the tube is no better than the point-edit baseline, the claim that stroke-based sculpting preserves brush fidelity fails for that profile.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that stroke-based sculpting of a neural SDF reduces to a fine-tuning problem with a well-chosen tubular sampling scheme. Given a stroke curve, the method defines a local coordinate frame with tangent, in-plane perpendicular, and surface-normal directions, then samples points both along the curve and across a perpendicular tubular neighborhood, projects them onto the original zero level set, and displaces each by a brush profile modulated along the stroke. Fine-tuning a SIREN MLP against these displaced target points, while holding a fixed set of Markovian surface samples for untouched regions, makes the updated zero level set coincide with the edited surface. The paper reports that this approach keeps edits interactive, preserves unedited surface regions better than repeated point edits, and supports custom brush profiles and stroke modulations that point-based editing cannot express cleanly.

Load-bearing premise

The load-bearing premise is that fine-tuning a small SIREN network with samples from the tubular stroke region plus a fixed set of 10,000 Markovian surface samples is enough to move the surface exactly where the brush says while leaving all unedited regions untouched.

Editorial extensions

If this is right

  • A user can carve or extrude along arbitrary curves on a neural SDF without converting to a mesh or point cloud.
  • Each stroke edit completes in under a second on current GPUs, making iterative, interactive sculpting feasible instead of a batch process.
  • Custom brush profiles and stroke modulation let one stroke produce chiseled edges, asymmetric bumps, scales, fading carves, or simultaneous carve-and-extrude effects.
  • Because the tubular sampler concentrates samples near the stroke, stroke edits need far fewer samples than sequential point edits, yielding up to a 16x speedup over the 3DNS baseline.

Reading between the lines

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

  • The paper leaves implicit that the same stroke sampling and profile displacement could be tested on other implicit field encoders, since its discussion notes the sculpting operator is agnostic to the underlying representation.
  • A natural extension the authors mention but do not develop is reversible editing: storing per-stroke parameter deltas would allow subtracting a stroke to undo it, which is essential for iterative design workflows.
  • The reported blurring of high-frequency brush profiles suggests a concrete benchmark for future backbones: strokes with several oscillations inside the brush radius should be used to compare whether a representation can faithfully reproduce the user's intended profile.
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

4 major / 5 minor

Summary. The paper presents INST-Sculpt, a framework for editing neural signed distance functions by drawing strokes. The method samples a tubular neighborhood around a user-defined curve, defines custom brush profiles and modulation functions, and fine-tunes a SIREN MLP to bring sampled points onto the target displaced surface. Experiments report Chamfer distance against mesh-based ground truth and compare editing time with the 3DNS point-edit baseline on six shapes, claiming lower Chamfer error and up to 16x speedup.

Significance. If the results hold, this is a useful step toward practical interactive sculpting of implicit neural representations. The paper extends point edits to stroke-based edits, introduces tubular sampling that substantially reduces interaction samples, and supports custom brush profiles and modulation, which are all valuable for graphics practice. The presentation is clear, and the direct comparison with 3DNS and with coarse-mesh editing is a concrete strength. However, the evaluation has important gaps: the ground truth is produced by the same brush formula used to generate training data, there are no error bars or significance tests, and the multi-stroke stability that the method's central claim depends on is not measured.

major comments (4)
  1. [Section 4.3, Table 2, Eq. (10)] The quantitative evaluation is self-referential: the ground-truth mesh edit is generated by applying the same brush displacement formula that is used to create the network's training samples. This measures how well the network can fit a target it was constructed to fit, rather than how well the method matches an independent sculpting intent. Please report per-edit distributions, add significance tests or confidence intervals, and validate against an independently defined ground truth such as edits produced by a mesh sculpting tool or by different deformation rules.
  2. [Section 5 and central claim] The paper claims that users can make stroke-based modifications 'on the fly' and 'intricate and versatile edits,' and Fig. 2 shows multi-stroke results, but no experiment measures cumulative drift or preservation of unedited regions after sequential strokes. Section 5 explicitly concedes that limited regularization causes bumpy artifacts while stronger regularization suppresses edits; this is precisely the localization/stability trade-off that the central claim depends on, and its operating point is not quantified. Please add a multi-stroke experiment with a control-region distance or drift metric, not just a per-edit average.
  3. [Section 3.5.1 and Table 1] The speedup comparison confounds algorithmic efficiency with sample-count reduction: 3DNS is run with 120,000 surface samples while INST-Sculpt uses 10,000, so the reported 'up to 16x speedup' largely reflects the smaller sample budget rather than an algorithmic advantage. Please compare at matched edit quality or matched Chamfer error, or report per-sample time and isolate the contribution of tubular sampling.
  4. [Section 5 and Fig. 7] The acknowledged blurring of high-frequency brush profiles limits the generality of the claim that 'intricate and versatile edits can be made while preserving the smooth nature of implicit representations.' This is not necessarily a fatal issue, but the paper should state it as a scope condition and ideally quantify the trade-off, for example by measuring error or fidelity as a function of brush oscillation frequency.
minor comments (5)
  1. [Section 4.3] There is a duplicated phrase in the text: 'we compare our method with three baselines:with three baselines:'.
  2. [Eq. (1)] The sign convention in Eq. (1) appears reversed relative to the standard SDF convention (negative inside, positive outside); please clarify or correct the definition.
  3. [Eq. (10)] In Eq. (10), n(u(p)) is described as 'the surface normal at the stroke,' but p is generally off the stroke; please clarify whether the normal is evaluated at the projected stroke point or at p itself.
  4. [Section 3.3] The brush profile is stated to be C1, but no continuity check is described at the boundaries v = ±1 or between the control points of the Catmull-Rom spline; please clarify the construction.
  5. [General] The paper does not state whether code or data will be released; a reproducibility statement would be helpful for a tools-oriented contribution.

Circularity Check

1 steps flagged · score 2.0 of 10

No structural circularity in the method; the only notable self-reference is that the quantitative ground truth for edits is generated by the same brush-displacement formula used to create the network's fine-tuning targets, making the tubular-region Chamfer numbers a fit-quality check rather than an external validation.

  1. fitted input called prediction [Sec. 3.3 Eq. (10)-(11) and Sec. 4.3 ground truth comparison]
    "p′ = p + Br,s(v(p)) n(u(p)), (10) ... The neural SDF is then fine-tuned to align its zero-level set with the deformed surface by minimizing: Ldeformation = E [|fθ(p′)|] , (11) ... For the ground truth comparison, we apply the brush profile to mesh vertices within a radius of curve segments, providing an ideal edit by directly deforming the mesh geometry."

    The fine-tuning supervision in Eq. (11) is applied at points p′ defined by the brush-displacement formula in Eq. (10), and the ground-truth mesh used for evaluation in Sec. 4.3 is produced by applying the same brush profile to mesh vertices. Consequently, the reported Chamfer distance, especially inside the tubular interaction region, measures how well the fine-tuned SIREN reproduces the exact formula that generated its training targets. The fit is statistically forced in the edited region: minimizing E[|f_theta(p′)|] at samples from the brush-displaced surface directly drives the network's zero level set toward that same surface. This is a self-referential benchmark rather than an external correctness test.

full rationale

The paper's core operation is a straightforward fine-tuning procedure: sample points in a tubular neighborhood of a user stroke, displace them by a user-defined brush profile, and minimize the SDF magnitude at those displaced points. There is no derivation chain in which a claimed result is obtained from premises that secretly contain the result; the method is explicitly a fitting procedure. The strongest self-reference is in the evaluation: the 'ground truth mesh edit' is generated by applying the same brush profile formula (Eq. 10) that creates the training targets (Eq. 11), so the tubular-region Chamfer numbers are a consistency check of the network's capacity and sampling density rather than evidence about the quality of the sculpting operation itself. This limits the interpretation of Table 2's absolute numbers, but does not undermine the paper's primary contribution, which is the stroke-based tubular sampling strategy enabling edits under a second. That contribution is independently supported by the timing comparison with 3DNS (Table 1) and by visual results. The paper also honestly states in Sec. 5 that limited regularization causes bumpy artifacts while stronger regularization suppresses edits, and that high-frequency brush profiles are blurred; this is a capacity limitation, not a circular argument. No self-citations are load-bearing: the loss and Markovian sampler are adopted from 3DNS, which is an external prior work by different authors, and the SIREN architecture is cited independently. Overall, the paper is not a circular derivation; it contains one self-referential evaluation design that mildly overstates the meaning of its Chamfer metric, consistent with a low score of 2.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical or mathematical entities. Its main dependencies are the SIREN backbone, the 3DNS loss and sampler (used as-is), and a set of hand-tuned sampling counts and brush parameters. The edit target is defined by the same brush formula that generates the ground truth, which keeps the circularity burden low but makes the evaluation somewhat self-referential.

free parameters (4)
  • sampling hyperparameters = 99 stroke samples x 101 offset samples, 10,000 surface samples
    Hand-chosen in Section 3.5.1; no sensitivity analysis except stroke counts in Table 3.
  • brush radius and intensity = r=0.08, s=0.06
    Experiment settings in Section 4.3; not fitted to data, but they define the evaluation target.
  • fine-tuning epochs = 100 for quality, 50 for timing
    Set per experiment in Sections 4.3 and 4.4; no ablation on epochs.
  • loss weights = λ1=1.5e3, λ2=5, λ3=2.5, λ4=5, α=100
    Adopted from 3DNS [28] with no re-tuning; they govern the balance between preserving the shape and applying the edit.
assumptions (5)
  • domain assumption The zero level set of the fine-tuned SIREN network coincides with the displaced sample points p' after minimizing E[|fθ(p')|].
    The method assumes that matching the SDF at sampled displaced points is sufficient to move the whole surface, treating the network as an implicit surface interpolator (Section 3.3, Eq. 11).
  • domain assumption The displacement p' = p + Br,s(v(p)) n(u(p)) defines a valid edit target and does not produce self-intersections or invalid geometry.
    No proof that normal offsets along a curved stroke yield a valid surface; standard practice in mesh sculpting, but relies on small offsets (Section 3.3, Eq. 10).
  • domain assumption The 3DNS Markovian SDF sampler produces a uniform and sufficient distribution of surface points for preserving unedited regions.
    Adopted without modification from [28] in Section 3.5.2; no analysis of coverage or failure cases.
  • domain assumption The SIREN architecture with sinusoidal activations can represent the intended edits well enough for interactive use.
    The authors note in Section 5 that high-frequency brush profiles are blurred by the network, so the choice of backbone bounds the achievable edit fidelity.
  • domain assumption Chamfer distance between the edited neural SDF's zero set and the mesh ground truth measures editing accuracy.
    The metric measures geometric closeness; it does not capture visual quality or whether the edit direction matches user intent (Section 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of INST-Sculpt: Interactive Stroke-based Neural SDF Sculpting." pith.science (2026). https://pith.science/paper/7WQMZP6R

@misc{pith2026250202891,
  author       = {Pith},
  title        = {Pith review of: INST-Sculpt: Interactive Stroke-based Neural SDF Sculpting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7WQMZP6R}},
  note         = {Machine review of arXiv:2502.02891}
}
read the original abstract

Recent advances in implicit neural representations have made them a popular choice for modeling 3D geometry, achieving impressive results in tasks such as shape representation, reconstruction, and learning priors. However, directly editing these representations poses challenges due to the complex relationship between model weights and surface regions they influence. Among such editing tools, sculpting, which allows users to interactively carve or extrude the surface, is a valuable editing operation to the graphics and modeling community. While traditional mesh-based tools like ZBrush facilitate fast and intuitive edits, a comparable toolkit for sculpting neural SDFs is currently lacking. We introduce a framework that enables interactive surface sculpting edits directly on neural implicit representations. Unlike previous works limited to spot edits, our approach allows users to perform stroke-based modifications on the fly, ensuring intuitive shape manipulation without switching representations. By employing tubular neighborhoods to sample strokes and custom brush profiles, we achieve smooth deformations along user-defined curves, providing precise control over the sculpting process. Our method demonstrates that intricate and versatile edits can be made while preserving the smooth nature of implicit representations.

Figures

Figures reproduced from arXiv: 2502.02891 by the authors.

Figure 1
Figure 1. Overview of our interactive neural sculpting pipeline. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Examples of natural artistic edits using our method. Each column shows a sample shape from the dataset, with the original SDF [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Examples of edits using different brush profiles and [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Effect of radius and intensity variations for the same [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the coordinate frame and sampling strat [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Impact of model resolution on brush template fidelity. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: Example of a stroke applied using different methods. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 31 canonical work pages

  1. [1]

    Sal: Sign agnos- tic learning of shapes from raw data

    Matan Atzmon and Yaron Lipman. Sal: Sign agnos- tic learning of shapes from raw data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2565–2574, 2020. 2, 3

    work page 2020

  2. [2]

    Sald: Sign agnostic learning with derivatives

    Matan Atzmon and Yaron Lipman. Sald: Sign agnostic learning with derivatives. arXiv preprint arXiv:2006.05400,

    arXiv 2006

  3. [3]

    Differentiable render- ing of neural sdfs through reparameterization

    Sai Praveen Bangaru, Michael Gharbi, Fujun Luan, Tzu-Mao Li, Kalyan Sunkavalli, Milos Hasan, Sai Bi, Zexiang Xu, Gilbert Bernstein, and Fredo Durand. Differentiable render- ing of neural sdfs through reparameterization. InSIGGRAPH Asia 2022 Conference Papers, pages 1–9, 2022. 3

    2022

  4. [4]

    Introduction to im- plicit surfaces

    Jules Bloomenthal and Chandrajit Bajaj. Introduction to im- plicit surfaces. Morgan Kaufmann, 1997. 1

    work page 1997

  5. [5]

    A class of local interpolating splines

    E Catmull. A class of local interpolating splines. Computer Aided Geometric Design/Academic Press, 1974. 5

    work page 1974

  6. [6]

    Deformed implicit field: Modeling 3d shapes with learned dense correspon- dence

    Yu Deng, Jiaolong Yang, and Xin Tong. Deformed implicit field: Modeling 3d shapes with learned dense correspon- dence. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 10286–10296,

    work page

  7. [7]

    MagicaCSG – A Nondestructive Signed Distance Field (SDF) Editor and Path Tracing Renderer

    Ephtracy. MagicaCSG – A Nondestructive Signed Distance Field (SDF) Editor and Path Tracing Renderer . Magi- caV oxel, 2023. 3

    work page 2023

  8. [8]

    12.7 constructive solid geometry

    James D Foley. 12.7 constructive solid geometry. Computer Graphics: Principles and Practice, pages 533–558, 1996. 1

    work page 1996

Show all 37 references
  1. [9]

    The fractal geometry of nature/revised and enlarged edition

    Benoit B Mandelbrot. The fractal geometry of nature/revised and enlarged edition. New York, 1983. 2

    1983

  2. [10]

    Zbrush – The industry standard for digital sculpting and painting

    Maxon. Zbrush – The industry standard for digital sculpting and painting. Maxon Computer GmbH, 2022. 3

    2022

  3. [11]

    A level set theory for neural implicit evolution under explicit flows

    Ishit Mehta, Manmohan Chandraker, and Ravi Ramamoor- thi. A level set theory for neural implicit evolution under explicit flows. In European Conference on Computer Vision, pages 711–729. Springer, 2022. 3

    2022

  4. [12]

    Occupancy networks: Learning 3d reconstruction in function space

    Lars Mescheder, Michael Oechsle, Michael Niemeyer, Se- bastian Nowozin, and Andreas Geiger. Occupancy networks: Learning 3d reconstruction in function space. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4460–4470, 2019. 3

    2019

  5. [13]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. Communications of the ACM, 65(1):99–106, 2021. 2, 3

    2021

  6. [14]

    Instant neural graphics primitives with a mul- tiresolution hash encoding

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a mul- tiresolution hash encoding. ACM transactions on graphics (TOG), 41(4):1–15, 2022. 3

    2022

  7. [15]

    Level set methods and dynamic implicit surfaces

    Stanley Osher, Ronald Fedkiw, and K Piechor. Level set methods and dynamic implicit surfaces. Appl. Mech. Rev., 57(3):B15–B15, 2004. 1

    2004

  8. [16]

    Deepsdf: Learning con- tinuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning con- tinuous signed distance functions for shape representation. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 165–174, 2019. 2, 3

    2019

  9. [17]

    Inigo Quilez – distance functions

    Inigo Quilez. Inigo Quilez – distance functions. 2

  10. [18]

    Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps

    Christian Reiser, Songyou Peng, Yiyi Liao, and Andreas Geiger. Kilonerf: Speeding up neural radiance fields with thousands of tiny mlps. In Proceedings of the IEEE/CVF international conference on computer vision , pages 14335– 14345, 2021. 3

    2021

  11. [19]

    Boolean op- erations in solid modeling: Boundary evaluation and merg- ing algorithms

    Aristides AG Requicha and Herbert B V oelcker. Boolean op- erations in solid modeling: Boundary evaluation and merg- ing algorithms. Proceedings of the IEEE, 73(1):30–44, 1985. 3

    1985

  12. [20]

    Direct manipulation of procedural implicit surfaces

    Marzia Riso, ´Elie Michel, Axel Paris, Valentin Deschaintre, Mathieu Gaillard, and Fabio Pellacini. Direct manipulation of procedural implicit surfaces. ACM Transaction on Graph- ics (SIGGRAPH Asia ’24 Conference Proceedings), 2024. 3

    2024

  13. [21]

    Blender - a 3D modelling and rendering package

    Ton Roosendaal. Blender - a 3D modelling and rendering package. Blender Foundation, 2018. 3

    2018

  14. [22]

    Free-form defor- mation of solid geometric models

    Thomas W Sederberg and Scott R Parry. Free-form defor- mation of solid geometric models. In Proceedings of the 13th annual conference on Computer graphics and interac- tive techniques, pages 151–160, 1986. 3

    1986

  15. [23]

    Implicit neural representa- tions with periodic activation functions

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representa- tions with periodic activation functions. Advances in neural information processing systems, 33:7462–7473, 2020. 1, 3, 4

    2020

  16. [24]

    Fourier features let networks learn high frequency functions in low dimen- sional domains

    Matthew Tancik, Pratul Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ra- mamoorthi, Jonathan Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimen- sional domains. Advances in neural information ...

    2020

  17. [25]

    Womp – The Goopiest, Easy to Use 3D Design Software

    Gabriela Trueba. Womp – The Goopiest, Easy to Use 3D Design Software. Womp 3D Inc, 2016. 3

    2016

  18. [26]

    TurboSquid 3D Models

    TurboSquid. TurboSquid 3D Models. 7

  19. [27]

    Zippered polygon meshes from range images

    Greg Turk and Marc Levoy. Zippered polygon meshes from range images. In Proceedings of the 21st annual conference on Computer graphics and interactive techniques , pages 311–318, 1994. 7

    1994

  20. [28]

    3d neural sculpting (3dns): Editing neural signed distance func- tions

    Petros Tzathas, Petros Maragos, and Anastasios Roussos. 3d neural sculpting (3dns): Editing neural signed distance func- tions. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 4521–4530, 2023. 2, 3, 5, 7

    2023

  21. [29]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 3

    2017

  22. [30]

    Computer-aided design applica- tions of the rational b-spline approximation form

    Kenneth James Versprille. Computer-aided design applica- tions of the rational b-spline approximation form. Syracuse University, 1975. 3

    1975

  23. [31]

    Vector field based shape deformations

    Wolfram V on Funck, Holger Theisel, and Hans-Peter Seidel. Vector field based shape deformations. ACM Transactions on Graphics (ToG), 25(3):1118–1125, 2006. 3

    2006

  24. [32]

    Neural fields in visual computing and beyond

    Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, Shiqin Yan, Numair Khan, Federico Tombari, James Tomp- kin, Vincent Sitzmann, and Srinath Sridhar. Neural fields in visual computing and beyond. Computer Graphics Forum,

  25. [33]

    Signal processing for implicit neural rep- resentations

    Dejia Xu, Peihao Wang, Yifan Jiang, Zhiwen Fan, and Zhangyang Wang. Signal processing for implicit neural rep- resentations. Advances in Neural Information Processing Systems, 35:13404–13418, 2022. 3

    2022

  26. [34]

    Geometry processing with neural fields

    Guandao Yang, Serge Belongie, Bharath Hariharan, and Vladlen Koltun. Geometry processing with neural fields. Advances in Neural Information Processing Systems , 34: 22483–22497, 2021. 3

    2021

  27. [35]

    Stabilizing the optimization of neural signed distance functions and finer shape representation

    Huizong Yang, Yuxin Sun, Ganesh Sundaramoorthi, and An- thony Yezzi. Stabilizing the optimization of neural signed distance functions and finer shape representation. Advances in Neural Information Processing Systems, 36, 2024. 3

    2024

  28. [36]

    Controllable shape modeling with neural generalized cylinder

    Xiangyu Zhu, Zhiqin Chen, Ruizhen Hu, and Xiaoguang Han. Controllable shape modeling with neural generalized cylinder. arXiv preprint arXiv:2410.03675, 2024. 3

    2024 arXiv

  29. [37]

    Interactive multiresolution mesh editing

    Denis Zorin, Peter Schr ¨oder, and Wim Sweldens. Interactive multiresolution mesh editing. In Proceedings of the 24th an- nual conference on Computer graphics and interactive tech- niques, pages 259–268, 1997. 3 10

    1997

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.