pith. sign in

arxiv: 2212.08790 · v2 · submitted 2022-12-17 · 💻 cs.GR

Unphased Wrinkles: Estimating cloth elasticity parameters using a frequency-based loss

Egor Larionov , Marie-Lena Eckert , Katja Wolff , Tuur Stuyck This is my paper

Pith reviewed 2026-05-24 10:26 UTC · model grok-4.3

classification 💻 cs.GR
keywords cloth parameter estimationdifferentiable simulationwrinkle frequency lossmaterial captureelasticity parameterstemplate registration
0
0 comments X

The pith

A frequency-based loss on wrinkled scans produces cloth elasticity parameters that stay consistent across different wrinkle patterns of the same fabric.

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

The paper tries to establish a simpler pipeline for recovering stretch and bend parameters from ordinary captures of real fabrics. Rather than designing experiments that separate stretch from bending, the method accepts that stretching produces wrinkles and compares the frequency content of those wrinkles between simulation and scan. An objective built on this comparison is shown to return similar parameter sets for multiple wrinkle configurations of one material. Bending is estimated first because membrane stiffness affects bending modes only weakly, after which differentiable simulation optimizes the full set once a template mesh has been registered to the captured surface. If the approach holds, material capture becomes feasible with everyday scanning and without specialized isolation rigs.

Core claim

By replacing direct geometric comparison with a frequency-domain loss on wrinkles and by estimating bending stiffness first, the optimization recovers elasticity parameters whose values depend on the fabric rather than on the particular deformed configuration; the same parameters are recovered from different wrinkle patterns once a template is registered to the scan.

What carries the argument

The frequency-based loss that measures spectral similarity of wrinkles between simulated and target cloth, applied after independent bending estimation and template registration to the scan.

If this is right

  • Parameters recovered this way can be reused for new deformations of the same fabric without re-estimation.
  • Capture hardware can be decoupled from the optimizer because only a registered template mesh is required.
  • Wrinkled data from ordinary scans become usable input instead of requiring stretch-only experiments.

Where Pith is reading between the lines

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

  • The same frequency comparison might be tested on other thin-sheet materials whose bending and stretching produce visible folds.
  • If the loss proves robust, it could be inserted into existing cloth simulators to tune parameters on the fly from video.
  • Extending the pipeline to time-varying captures would let the method handle dynamic wrinkling without new machinery.

Load-bearing premise

Bending stiffness can be estimated first without membrane stiffness interfering, and the frequency loss returns values that are specific to the material rather than to any given wrinkle pattern.

What would settle it

Optimize the parameters on two visibly different wrinkle configurations of the identical physical fabric and check whether the recovered stretch and bend values agree within the precision of the capture and simulation.

Figures

Figures reproduced from arXiv: 2212.08790 by Egor Larionov, Katja Wolff, Marie-Lena Eckert, Tuur Stuyck.

Figure 1
Figure 1. Figure 1: Parameter estimation pipeline. With our pipeline, we decouple the cloth capture (left) from the parameter optimization (middle-right) using NR-ICP mesh registration (middle-left). The optimization pass is able to handle wrinkled cloth, which greatly simplifies the capture process. With the optimized parameters 𝛾, we generate realistic full-body cloth simulation (right) bypassing laborious manual parameter … view at source ↗
Figure 2
Figure 2. Figure 2: Cloth capture. Our simple cloth capture system records the cloth swatch under different force applications to drive the optimization. et al. 2013; Macklin et al. 2016; Müller et al. 2007; Overby et al. 2017; Stuyck 2018]. In summary, the research field has made great progress while several issues still remain. Deeply intertwined pipelines of closely coupled capture and optimization systems complicate the c… view at source ↗
Figure 3
Figure 3. Figure 3: Registration. Capture setup (left), scanned mesh (middle), and regis￾tered mesh (right) with magnified regions in the insets. with the stamped grid pattern are projected along the mesh normal towards the surface of the scanned mesh to improve the position estimate in the normal direction. While more accurate methods have been developed for establishing dense correspondences between meshes, we stress that a… view at source ↗
Figure 4
Figure 4. Figure 4: Wrinkle bifurcation: two valid cloth configurations with contrary wrinkle patterns shown from above (top) and side (bottom). The same cloth swatch is pulled with the same force at opposite corners. As outlined above, the evaluation of the objective in Eq. 8 re￾sults in a very different value for both equilibrium states in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Shape descriptor evaluation. Target cloth is simulated with XPBD (left) followed by optimization results for each descriptor (right). Euclidean vertex distances to the target are color-coded with a maximum of 1.8 mm. Cotton Denim Silk [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distinct equilibrium states for the same material. Two simulations for each material with different initial conditions: perturbed vertices (left), flat configuration (middle), and color-coded Euclidean distance between both meshes in red (right), with a maximum distance of 4.3 mm [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Relative metric differences. Values of Δ𝑠ˆpos,𝑚,𝑚2,𝑖, 𝑗 (left) and Δ𝑠ˆFFT,𝑚,𝑚2,𝑖, 𝑗 (right) where 𝑚 is the 𝑥-axis: spos doesn’t necessarily produce higher metric differences across materials compared to varying cloth config￾urations for the same material, as shown by the negative and close-to-zero values (left), while sFFT does as observed in the mostly positive values (right). eight instances of Δ𝑠ˆ𝐹 𝐹𝑇 ,… view at source ↗
Figure 8
Figure 8. Figure 8: Here, we choose high stiffness values due to finding the simulation [PITH_FULL_IMAGE:figures/full_fig_p005_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Re-simulation with the estimation of Houdini’s silk preset. The vertex displacement error is color-coded in red with a maximum set to 2 mm. In the second column, errors are clustered on the large stretch areas with a maximum of 6 mm while it is below 2.4 mm in the remaining columns. all deformations apart from the second column, the local vertex displacement is below 2.4 mm. In the second column, the erro… view at source ↗
Figure 9
Figure 9. Figure 9: Re-simulation of various cloth configurations with 𝛾FFT,c estimated from XPBD targets with 𝛾cotton. The vertex displacement error is color-coded in red, with a small maximum error of 0.28 mm. The second and last columns are targets while the rest are unseen by the optimization. 5 RESULTS Synthetic Experiments. We show that our method is capable of re￾producing cloth simulations from third-party software. F… view at source ↗
Figure 12
Figure 12. Figure 12: Aesthetic evaluation. By twisting a piece of cloth, unseen by op￾timization, we demonstrate that distinct wrinkle patterns generated by Houdini’s FEM simulator are reproduced using our XPBD simulator with optimized material parameters [PITH_FULL_IMAGE:figures/full_fig_p006_12.png] view at source ↗
Figure 15
Figure 15. Figure 15: Outfit simulation: a denim overall and cotton top (left) and a polyester soccer outfit (right). 6 LIMITATIONS AND FUTURE WORK We presented a novel approach to cloth captures and optimization that simplifies the inverse design problem of parameter estimation. Despite improving robustness and simplicity over prior work, sev￾eral limitations remain. For instance, the accuracy of our template registration tec… view at source ↗
Figure 14
Figure 14. Figure 14: Real-world results. Denim (top), cotton (middle), and polyester (bot￾tom) cloth is simulated with our estimated parameters. The vertex position error to the targets in [PITH_FULL_IMAGE:figures/full_fig_p007_14.png] view at source ↗
read the original abstract

Generating realistic clothing for virtual applications like online retail and digital avatars is crucial but requires expert knowledge of 3D tools to generating believable simulations. Recently, a number of works proposed to estimate cloth material properties from specialized capture setups. However, these systems tend to be monolithic, complex and expensive. We propose a simplified method for automatically determining parameters based on easily captured real-world fabrics. While existing methods carefully design experiments to isolate stretch parameters from bending modes, we embrace that stretching fabrics causes wrinkling and propose a novel specialized loss for comparing wrinkled fabrics. We designed our objective function to capture material-specific behavior, resulting in similar values for different wrinkle configurations of the same material. We estimate bending first, given that membrane stiffness has little effect on bending. We use differentiable simulation to find an optimal set of parameters that minimizes the difference between simulated cloth and deformed target cloth. Furthermore, our pipeline decouples the capture method from the optimization by registering a template mesh to the scanned data. These choices simplify the capture system and allow for wrinkles in scanned fabrics. We demonstrate our method on captured data of three different real-world fabrics and on three digital fabrics produced by a third-party simulator.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a simplified pipeline for estimating cloth bending and stretch parameters from easily captured real-world fabrics that exhibit wrinkles. It uses differentiable simulation with a frequency-based loss to compare simulated and target deformed cloth, registers a template mesh to decouple capture from optimization, and performs a two-stage procedure: first optimizing bending stiffness while holding membrane parameters fixed (justified by the claim that membrane stiffness has little effect on bending modes), then optimizing stretch parameters. The authors assert that the objective function captures material-specific behavior, yielding similar parameter values across different wrinkle configurations of the same fabric, and demonstrate the approach on three real fabrics plus three digital ones from a third-party simulator.

Significance. If the frequency-based loss produces configuration-independent parameters and the two-stage decoupling is valid, the method would offer a lower-barrier alternative to existing specialized capture systems for cloth material estimation in graphics applications such as virtual avatars and online retail. The explicit handling of wrinkles rather than avoiding them is a practical strength, and the use of template registration improves flexibility.

major comments (2)
  1. [method description (bending estimation procedure)] Method description (bending-first stage): The justification for estimating bending parameters first while holding membrane stiffness fixed rests on the statement that 'membrane stiffness has little effect on bending,' but no ablation or sensitivity analysis quantifies how much the recovered bending values shift when membrane parameters are varied by even one order of magnitude. In wrinkling regimes, membrane resistance directly governs wrinkle wavelength and amplitude, so the deformed shape seen by the bending optimizer is not independent of membrane stiffness; this assumption is load-bearing for the claim of isolating material-specific bending parameters.
  2. [results and objective function claims] Results and claims of configuration independence: The assertion that the frequency-based loss 'resulting in similar values for different wrinkle configurations of the same material' is central, yet the manuscript provides no quantitative metric (variance, standard deviation, or statistical comparison) of parameter consistency across configurations, nor an ablation showing that the loss remains stable when membrane stiffness is altered. Without this, it is unclear whether the reported similarity is robust or an artifact of the specific captured setups.
minor comments (2)
  1. [abstract and method] The abstract and method sections would benefit from a brief equation or pseudocode sketch of the frequency-based loss to clarify how frequency content is extracted and compared.
  2. [optimization details] Clarify the exact number and type of parameters optimized in each stage (e.g., which stretch parameters are included) and whether any regularization is applied.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our method and results. The comments correctly identify places where additional quantitative support would strengthen the manuscript. We address each major comment below and commit to revisions that incorporate the requested analyses.

read point-by-point responses

  1. Referee: Method description (bending estimation procedure): The justification for estimating bending parameters first while holding membrane stiffness fixed rests on the statement that 'membrane stiffness has little effect on bending,' but no ablation or sensitivity analysis quantifies how much the recovered bending values shift when membrane parameters are varied by even one order of magnitude. In wrinkling regimes, membrane resistance directly governs wrinkle wavelength and amplitude, so the deformed shape seen by the bending optimizer is not independent of membrane stiffness; this assumption is load-bearing for the claim of isolating material-specific bending parameters.

    Authors: We agree that the two-stage procedure would be better supported by explicit quantification. In the revised manuscript we will add a sensitivity analysis that varies membrane stiffness parameters over at least one order of magnitude while re-running the bending optimization stage, reporting the resulting shifts in recovered bending values. This will directly address the concern that the deformed shape is not independent of membrane stiffness. revision: yes

  2. Referee: Results and claims of configuration independence: The assertion that the frequency-based loss 'resulting in similar values for different wrinkle configurations of the same material' is central, yet the manuscript provides no quantitative metric (variance, standard deviation, or statistical comparison) of parameter consistency across configurations, nor an ablation showing that the loss remains stable when membrane stiffness is altered. Without this, it is unclear whether the reported similarity is robust or an artifact of the specific captured setups.

    Authors: We acknowledge that the claim of configuration-independent parameters requires quantitative backing. The revised version will include (1) a table reporting mean, variance, and standard deviation of the estimated parameters across the different wrinkle configurations for each fabric, and (2) an ablation that perturbs membrane stiffness and measures stability of the frequency loss and recovered parameters. These additions will supply the requested statistical comparison. revision: yes

Circularity Check

0 steps flagged

No circularity: parameter fitting to external capture data

full rationale

The paper describes an optimization pipeline that fits bending and membrane parameters to registered real-world capture data using a frequency-based loss and differentiable simulation. The central claim is that the recovered parameters are material-specific and configuration-independent; this is an empirical outcome of the fitting procedure rather than a mathematical reduction in which a derived quantity equals its own input by construction. The stated decoupling assumption ('membrane stiffness has little effect on bending') is an external modeling choice, not a self-referential definition or a fitted input relabeled as a prediction. No self-citations, uniqueness theorems, or ansatzes imported from prior author work appear in the derivation chain. The method is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

Abstract-only view yields minimal ledger entries; full paper would be needed to enumerate all fitted parameters and background assumptions.

free parameters (2)
  • bending stiffness
    Estimated first via optimization; value fitted to match bending behavior of captured data.
  • stretch parameters
    Estimated after bending; values fitted to minimize frequency loss on wrinkled configurations.
axioms (2)
  • domain assumption Differentiable cloth simulation accurately reproduces real wrinkling behavior under the chosen constitutive model.
    Central to the optimization loop that finds parameters matching captured data.
  • domain assumption Membrane stiffness has negligible influence on pure bending modes.
    Justifies sequential estimation of bending before stretch parameters.

pith-pipeline@v0.9.0 · 5748 in / 1329 out tokens · 22190 ms · 2026-05-24T10:26:15.926903+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

  1. [1]

    Eurographics Association

    In EUROGRAPHICS 2017 Tutorials. Eurographics Association. Kiran S. Bhat, Christopher D. Twigg, Jessica K. Hodgins, Pradeep K. Khosla, Zoran Popović, and Steven M. Seitz

    work page 2017

  2. [2]

    In Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Com- puter Animation (San Diego, California) (SCA ’03)

    Estimating Cloth Simulation Parameters from Video. In Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Com- puter Animation (San Diego, California) (SCA ’03). Eurographics Association, Goslar, DEU, 37–51. Sofien Bouaziz, Sebastian Martin, Tiantian Liu, Ladislav Kavan, and Mark Pauly

    work page 2003

  3. [3]

    ACM trans- actions on graphics (TOG) 33, 4 (2014), 1–11

    Projective dynamics: Fusing constraint projections for fast simulation. ACM trans- actions on graphics (TOG) 33, 4 (2014), 1–11. Browzwear

    work page 2014

  4. [4]

    arXiv preprint arXiv:2201.04623 (2022)

    Virtual Elastic Objects. arXiv preprint arXiv:2201.04623 (2022). CLO 3D

    work page arXiv 2022

  5. [5]

    In Proceedings of the ACM SIGGRAPH / Eurographics Symposium on Computer Animation (Los Angeles, California) (SCA ’17)

    Modeling and Data-Driven Parameter Estimation for Woven Fabrics. In Proceedings of the ACM SIGGRAPH / Eurographics Symposium on Computer Animation (Los Angeles, California) (SCA ’17). Association for Computing Machinery, New York, NY, USA, Article 17, 11 pages. https://doi.org/10.1145/3099564.3099577 Tao Du, Kui Wu, Pingchuan Ma, Sebastien Wah, Andrew Spi...

    work page doi:10.1145/3099564.3099577

  6. [6]

    arXiv preprint arXiv:2101.05917 (2021)

    DiffPD: Differentiable Projective Dynamics with Contact. arXiv preprint arXiv:2101.05917 (2021). Marco Fratarcangeli, Valentina Tibaldo, and Fabio Pellacini

    work page arXiv 2021

  7. [7]

    ACM Transactions on Graphics (TOG) 35, 6 (2016), 1–9

    Vivace: A practical gauss-seidel method for stable soft body dynamics. ACM Transactions on Graphics (TOG) 35, 6 (2016), 1–9. Carsten Griwodz, Simone Gasparini, Lilian Calvet, Pierre Gurdjos, Fabien Castan, Benoit Maujean, Gregoire De Lillo, and Yann Lanthony

    work page 2016

  8. [8]

    In Proceedings of the 12th ACM Multimedia Systems Conference - MMSys ’21

    AliceVision Meshroom: An open-source 3D reconstruction pipeline. In Proceedings of the 12th ACM Multimedia Systems Conference - MMSys ’21 . ACM Press. https://doi.org/10.1145/3458305. 3478443 Yuanming Hu, Jiancheng Liu, Andrew Spielberg, Joshua B Tenenbaum, William T Freeman, Jiajun Wu, Daniela Rus, and Wojciech Matusik

    work page doi:10.1145/3458305

  9. [9]

    Proceedings of IEEE International Conference on Robotics and Automation (ICRA) (2019)

    ChainQueen: A Real-Time Differentiable Physical Simulator for Soft Robotics. Proceedings of IEEE International Conference on Robotics and Automation (ICRA) (2019). Krishna Murthy Jatavallabhula, Miles Macklin, Florian Golemo, Vikram Voleti, Linda Petrini, Martin Weiss, Breandan Considine, Jerome Parent-Levesque, Kevin Xie, Kenny Erleben, Liam Paull, Flori...

    work page 2019

  10. [10]

    International Conference on Learning Representations (ICLR) (2021)

    gradSim: Differentiable simulation for system identification and visuomotor control. International Conference on Learning Representations (ICLR) (2021). https: //openreview.net/forum?id=c_E8kFWfhp0 Eunjung Ju and Myung Geol Choi

    work page 2021

  11. [11]

    IEEE Access 8 (2020), 195113–195121

    Estimating Cloth Simulation Parameters From a Static Drape Using Neural Networks. IEEE Access 8 (2020), 195113–195121. Hao Li, Robert W Sumner, and Mark Pauly

    work page 2020

  12. [12]

    https://proceedings.neurips.cc/paper/2019/file/ 28f0b864598a1291557bed248a998d4e-Paper.pdf Tiantian Liu, Adam W Bargteil, James F O’Brien, and Ladislav Kavan

    Curran Associates, Inc. https://proceedings.neurips.cc/paper/2019/file/ 28f0b864598a1291557bed248a998d4e-Paper.pdf Tiantian Liu, Adam W Bargteil, James F O’Brien, and Ladislav Kavan

    work page 2019

  13. [13]

    ACM Transactions on Graphics (TOG) 32, 6 (2013), 1–7

    Fast simulation of mass-spring systems. ACM Transactions on Graphics (TOG) 32, 6 (2013), 1–7. Christiane Luible

    work page 2013

  14. [14]

    Computer Graphics Forum 31, 2pt2 (2012), 519–528

    Data-Driven Estimation of Cloth Simulation Models. Computer Graphics Forum 31, 2pt2 (2012), 519–528. https://doi.org/10.1111/j.1467- 8659.2012.03031.x arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1467- 8659.2012.03031.x Matthias Müller, Bruno Heidelberger, Marcus Hennix, and John Ratcliff

    work page doi:10.1111/j.1467- 2012

  15. [15]

    Journal of Visual Communication and Image Representation 18, 2 (2007), 109–118

    Position based dynamics. Journal of Visual Communication and Image Representation 18, 2 (2007), 109–118. Optitex

    work page 2007

  16. [16]

    Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, and Ming C

    IEEE Transactions on Visualization and Computer Graphics 23, 10 (2017), 2222–2234. Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, and Ming C. Lin

    work page 2017

  17. [17]

    Synthesis Lectures on Visual Computing: Computer Graphics, Animation, Computational Photography, and Imaging 10, 3 (2018), 1–121

    Cloth Simulation for Computer Graphics. Synthesis Lectures on Visual Computing: Computer Graphics, Animation, Computational Photography, and Imaging 10, 3 (2018), 1–121. Huamin Wang, James F O’Brien, and Ravi Ramamoorthi

    work page 2018

  18. [18]

    ACM transactions on graphics (TOG) 30, 4 (2011), 1–12

    Data-driven elastic models for cloth: modeling and measurement. ACM transactions on graphics (TOG) 30, 4 (2011), 1–12. S. Weiss, R. Maier, D. Cremers, R. Westermann, and N. Thuerey

    work page 2011