pith. sign in

arxiv: 2606.06887 · v1 · pith:W654QEYBnew · submitted 2026-06-05 · 💻 cs.CV

ARAPDiffusion: ARAP Regularization for Diffusion-Based Deformable Shape Space Learning

Haibo Liu , Jinghan Ke , Haitao Yang , Xiangru Huang , Georgios Pavlakos , Qixing Huang This is my paper

Pith reviewed 2026-06-27 22:52 UTC · model grok-4.3

classification 💻 cs.CV
keywords latent diffusionARAP regularizationdeformable shape generationpoint cloud shapesshape space learninggenerative models3D synthesis
0
0 comments X

The pith

ARAP regularization in latent diffusion reduces the need for large 3D datasets when learning deformable shape spaces.

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

The paper introduces ARAPDiffusion, a latent diffusion model that learns the continuous shape space of a collection of deformable shapes. It injects the as-rigid-as-possible deformation model as regularization losses into the diffusion process. This setup allows training without abundant 3D data by alternating between using the diffusion model's synthetic distribution to regularize the shape encoder and decoder, and using the decoder to regularize the diffusion model. The approach pairs a representation-free diffusion process with an implicit decoder suited to unorganized point clouds. Experiments on unconditional and conditional generation tasks show gains over standard latent diffusion baselines.

Core claim

ARAPDiffusion injects the as-rigid-as-possible deformation model as regularization losses into latent diffusion, with an alternating training procedure that uses the synthetic distribution from the diffusion model to improve the shape encoder and decoder and then uses the improved decoder to refine the diffusion model, thereby learning continuous shape spaces from limited 3D data while supporting implicit decoding of unorganized point clouds.

What carries the argument

The ARAP deformation model used as regularization losses inside the latent diffusion objective, combined with the alternating optimization loop between the diffusion model and the shape encoder/decoder.

If this is right

  • The method generates shapes from limited training collections where standard latent diffusion fails due to data scarcity.
  • The implicit decoder allows direct application to unorganized point clouds without explicit surface representations.
  • Both the encoder/decoder and the diffusion model improve through the mutual regularization loop.
  • Unconditional and conditional generation tasks benefit from the ARAP constraint in the learned shape space.

Where Pith is reading between the lines

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

  • The alternating regularization pattern could be tested with other local deformation energies beyond ARAP to handle different classes of shapes.
  • The representation-free diffusion step might combine with other implicit decoders for tasks such as shape interpolation under physical constraints.
  • If the convergence assumption holds, the same loop could be applied to conditional generation from partial observations or images.

Load-bearing premise

The alternating training between the latent diffusion model and the shape decoder will converge to mutual improvement rather than a degenerate fixed point.

What would settle it

Run the alternating procedure on a small collection of deformable shapes and check whether generated shapes violate ARAP energy bounds more than a non-regularized latent diffusion baseline trained on the same data.

Figures

Figures reproduced from arXiv: 2606.06887 by Georgios Pavlakos, Haibo Liu, Haitao Yang, Jinghan Ke, Qixing Huang, Xiangru Huang.

Figure 1
Figure 1. Figure 1: (Top) Intermediate shapes from ARAPReg [21], which exhibit geometric distortions. (Bottom) Intermediate shapes from ARAPDiffusion. ARAPDiffusion proceeds in three stages. The first stage pre-trains a generic LD model merely using the training data. The second stage fine-tunes the decoder by enforcing an ARAP regularization loss using the distribution of the latent space defined by the current LD model. The… view at source ↗
Figure 2
Figure 2. Figure 2: ARAPDiffusion has three stages. where stage II and stage III are alternated. (Left) Stage [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Correlation between our latent code scoring function and the reconstruction errors using [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Conditional generation results. The recon [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparisons between ARAPDiffusion and baseline approaches on unconditional implicit shape generation. ARAPDiffusion achieves better reconstruc￾tion errors. NoDeReg. In the second ablation study, we remove the decoder regularization term. In other words, we fix the decoder and use the scoring function w(z) to weight each latent code for fine-tuning the LD model. This variant is better than NoReg as the diff… view at source ↗
Figure 6
Figure 6. Figure 6: Color-coded reconstruction errors on test samples. We compare ARAPDiffusion with top [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Convergence of alternating optimization of ARAPDiffusion under different settings: (a) and [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: ARAPReg results. Red boxes indicate global failures; blue boxes indicate local artifacts. [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FrameAVE results. They show improved individual shape quality and higher success rate. [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: GeoLatent results. They show improved individual shape quality and higher success rate. [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: BRESA results. They show improved individual shape quality and higher success rate. [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: ARAPDiffusion results. They show improved individual shape quality and higher success [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The closest training shape of each generated shape. In each cell, the generated shape is [PITH_FULL_IMAGE:figures/full_fig_p023_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: ARAPReg results. Similar to other baseline approaches, some of the generated shapes [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FrameAVE results. Similar to other baseline approaches, some of the generated shapes [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: GeoLatent results. Similar to other baseline approaches, some of the generated shapes [PITH_FULL_IMAGE:figures/full_fig_p024_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: BREAS results. Similar to other baseline approaches, some of the generated shapes have [PITH_FULL_IMAGE:figures/full_fig_p025_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: ARAPDiffusion results. They show improved individual shape quality and higher success [PITH_FULL_IMAGE:figures/full_fig_p025_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: The closest training shape of each generated shape. In each cell, the generated shape is [PITH_FULL_IMAGE:figures/full_fig_p025_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Human conditional generation results. Our results align with the ground-truth better than [PITH_FULL_IMAGE:figures/full_fig_p026_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Animal conditional generation results. Our results align with the ground-truth better than [PITH_FULL_IMAGE:figures/full_fig_p027_21.png] view at source ↗
read the original abstract

This paper introduces ARAPDiffusion, a latent diffusion model to learn the underlying continuous shape space of a deformation shape collection. The key innovation is in injecting the as-rigid-as-possible (ARAP) deformation model as regularization losses into latent diffusion (LD), releasing the requirement of having abundant 3D training data for learning generative models. In contrast to the standard LD, we show how the ARAP model can be used to improve both the encoder/decoder and the LD model. The training procedure alternates between using the synthetic distribution defined by the LD model to develop a regularization loss that enhances the shape encoder/decoder and using the shape decoder to develop a regularization loss to improve the LD model. We also show the benefit of the LD paradigm in combining a representation-free LD process and an implicit shape decoder that is applicable to unorganized point clouds. The experimental results of unconditional and conditional shape generation demonstrate the advantages of ARAPDiffusion over baseline approaches.

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

1 major / 1 minor

Summary. The paper introduces ARAPDiffusion, a latent diffusion model for learning the continuous shape space of deformable shape collections. The core idea is to inject the as-rigid-as-possible (ARAP) deformation model as regularization losses into the latent diffusion process, thereby reducing the need for abundant 3D training data. Training alternates between (i) using synthetic samples from the latent diffusion model to regularize the shape encoder/decoder via ARAP losses and (ii) using the decoder to provide ARAP-based regularization for the diffusion model. The approach combines a representation-free diffusion process with an implicit decoder applicable to unorganized point clouds and reports advantages for unconditional and conditional shape generation over baselines.

Significance. If the alternating ARAP-regularized procedure can be shown to produce stable, non-degenerate improvement rather than collapse, the method would offer a practical route to generative shape modeling under limited 3D data by leveraging a classical geometric prior; this would be of clear interest to the computer-graphics and 3D-vision communities.

major comments (1)
  1. [Abstract] Abstract (training procedure paragraph): the central claim that ARAP regularization 'releases the requirement of having abundant 3D training data' rests on the alternating loop between the latent diffusion synthetic distribution and the ARAP-regularized decoder producing joint improvement. No convergence analysis, damping mechanism, or fixed-point stability argument is supplied; the procedure is formally analogous to an unregularized alternating game and could converge to a trivial solution (e.g., decoder outputs only near-rigid shapes while the diffusion model matches that narrow distribution) while still satisfying the per-step losses.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'representation-free LD process' is used without definition; it is unclear whether this refers to the latent space, the diffusion process itself, or the decoder architecture.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed feedback on the alternating training procedure. We address the concern regarding potential instability or collapse below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (training procedure paragraph): the central claim that ARAP regularization 'releases the requirement of having abundant 3D training data' rests on the alternating loop between the latent diffusion synthetic distribution and the ARAP-regularized decoder producing joint improvement. No convergence analysis, damping mechanism, or fixed-point stability argument is supplied; the procedure is formally analogous to an unregularized alternating game and could converge to a trivial solution (e.g., decoder outputs only near-rigid shapes while the diffusion model matches that narrow distribution) while still satisfying the per-step losses.

    Authors: We acknowledge that the manuscript does not include a formal convergence analysis, damping mechanism, or fixed-point stability argument for the alternating optimization. The referee is correct that such an analysis is absent. However, the procedure is not an unregularized alternating game: both the encoder/decoder and diffusion model are optimized with a combination of reconstruction losses on the limited real data, ARAP regularization terms that explicitly penalize non-rigid deviations while preserving local rigidity, and the latent diffusion objective that matches the empirical data distribution. This combination discourages collapse to a narrow near-rigid distribution, as the real training shapes contain non-rigid deformations. Our experiments (Sections 4.1–4.3) demonstrate that generated shapes maintain diversity and outperform baselines on metrics such as coverage and MMD without evident degeneracy. We will revise the paper to add an explicit discussion of these empirical safeguards and the role of the combined losses in preventing trivial solutions. revision: partial

Circularity Check

0 steps flagged

No significant circularity; alternating ARAP regularization is externally grounded

full rationale

The provided abstract and description present ARAP as an external deformation prior injected via alternating losses between LD synthetic samples and the shape decoder. No equations, fitted parameters renamed as predictions, or self-citation chains are visible that would reduce the central claim (ARAP releases need for abundant 3D data) to the inputs by construction. The procedure is described as a methodological choice rather than a self-definitional loop, and no uniqueness theorems or ansatzes from prior author work are invoked. The derivation remains self-contained against standard LD and ARAP benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review is abstract-only; no explicit free parameters, invented entities, or additional axioms beyond standard diffusion training are stated.

axioms (1)
  • domain assumption ARAP deformation energy supplies a useful inductive bias for regularizing both the shape autoencoder and the latent diffusion process on deformable collections.
    Invoked as the key mechanism that allows training with limited data.

pith-pipeline@v0.9.1-grok · 5712 in / 1281 out tokens · 18674 ms · 2026-06-27T22:52:14.208674+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

57 extracted references · 3 canonical work pages · 1 internal anchor

  1. [1]

    K. S. Arun, T. S. Huang, and S. D. Blostein. Least-squares fitting of two 3-d point sets.IEEE Transactions on Pattern Analysis and Machine Intelligence, 9(1):67–71, 1987

    1987

  2. [2]

    Atzmon, J

    M. Atzmon, J. Huang, F. Williams, and O. Litany. Approximately piecewise E(3) equivariant point networks. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024

    2024

  3. [3]

    Atzmon and Y

    M. Atzmon and Y . Lipman. SALD: sign agnostic learning with derivatives. In9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021

    2021

  4. [4]

    Atzmon, K

    M. Atzmon, K. Nagano, S. Fidler, S. Khamis, and Y . Lipman. Frame averaging for equivariant shape space learning. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 621–631. IEEE, 2022

    2022

  5. [5]

    Atzmon, K

    M. Atzmon, K. Nagano, S. Fidler, S. Khamis, and Y . Lipman. Frame averaging for equivariant shape space learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 631–641, June 2022

    2022

  6. [6]

    Atzmon, D

    M. Atzmon, D. Novotný, A. Vedaldi, and Y . Lipman. Augmenting implicit neural shape representations with explicit deformation fields.CoRR, abs/2108.08931, 2021

    work page arXiv 2021

  7. [7]

    Bouritsas, S

    G. Bouritsas, S. Bokhnyak, S. Ploumpis, S. Zafeiriou, and M. M. Bronstein. Neural 3d morphable models: Spiral convolutional networks for 3d shape representation learning and generation. In2019 IEEE/CVF International Conference on Computer Vision, ICCV 2019, Seoul, Korea (South), October 27 - November 2, 2019, pages 7212–7221. IEEE, 2019

    2019

  8. [8]

    Cheng, H.-Y

    Y .-C. Cheng, H.-Y . Lee, S. Tulyakov, A. G. Schwing, and L.-Y . Gui. Sdfusion: Multimodal 3d shape completion, reconstruction, and generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4456–4465, 2023

    2023

  9. [9]

    G. Chou, Y . Bahat, and F. Heide. Diffusion-sdf: Conditional generative modeling of signed distance functions. InProceedings of the IEEE International Conference on Computer Vision (ICCV), 2023

    2023

  10. [10]

    Dummer, N

    S. Dummer, N. Strisciuglio, and C. Brune. Rda-inr: Riemannian diffeomorphic autoencoding via implicit neural representations.SIAM Journal on Imaging Sciences, 17(4):2302–2330, 2024

    2024

  11. [11]

    Eisenberger, D

    M. Eisenberger, D. Novotný, G. Kerchenbaum, P. Labatut, N. Neverova, D. Cremers, and A. Vedaldi. Neuromorph: Unsupervised shape interpolation and correspondence in one go. InIEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021, pages 7473–7483. Computer Vision Foundation / IEEE, 2021

    2021

  12. [12]

    S. Foti, B. Koo, D. Stoyanov, and M. J. Clarkson. 3d generative model latent disentanglement via local eigenprojection.Comput. Graph. Forum, 42(6), 2023

    2023

  13. [13]

    Hartman, N

    E. Hartman, N. Charon, and M. Bauer. Self supervised networks for learning latent space representations of human body scans and motions, 2024

    2024

  14. [14]

    Hartman, E

    E. Hartman, E. Pierson, M. Bauer, N. Charon, and M. Daoudi. Bare-esa: A riemannian framework for unregistered human body shapes. InIEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, October 1-6, 2023, pages 14135–14145. IEEE, 2023

    2023

  15. [15]

    Hartman, E

    E. Hartman, E. Pierson, M. Bauer, M. Daoudi, and N. Charon. Basis restricted elastic shape analysis on the space of unregistered surfaces.Int. J. Comput. Vis., 133(4):1999–2024, 2025. 10

    1999

  16. [16]

    Hartman, Y

    E. Hartman, Y . Sukurdeep, E. Klassen, N. Charon, and M. Bauer. Elastic shape analysis of surfaces with second-order sobolev metrics: A comprehensive numerical framework.Int. J. Comput. Vis., 131(5):1183–1209, 2023

    2023

  17. [17]

    H.-I. Ho, C. Guo, P.-C. Wu, I. Shugurov, C. Tang, A. Mittal, S. An, M. Kaufmann, and L. Zhang. Phd: Personalized 3d human body fitting with point diffusion. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 7526–7537, 2025

    2025

  18. [18]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models. In H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, editors,Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, 2020

    2020

  19. [19]

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen. Lora: Low-rank adaptation of large language models. InThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022

    2022

  20. [20]

    Huang and D

    J. Huang and D. Mumford. Statistics of natural images and models. In1999 Conference on Computer Vision and Pattern Recognition (CVPR ’99), 23-25 June 1999, Ft. Collins, CO, USA, pages 1541–1547. IEEE Computer Society, 1999

    1999

  21. [21]

    Huang, X

    Q. Huang, X. Huang, B. Sun, Z. Zhang, J. Jiang, and C. Bajaj. Arapreg: An as-rigid-as possible regularization loss for learning deformable shape generators. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 5815–5825, October 2021

    2021

  22. [22]

    Karras, M

    T. Karras, M. Aittala, T. Aila, and S. Laine. Elucidating the design space of diffusion-based generative models. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors,Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28...

    2022

  23. [23]

    M. Kass, A. Witkin, and D. Terzopoulos. Snakes: Active contour models.International Journal of Computer Vision, 1(4):321–331, 1988

    1988

  24. [24]

    M. Kass, A. P. Witkin, and D. Terzopoulos. Snakes: Active contour models.Int. J. Comput. Vis., 1(4):321–331, 1988

    1988

  25. [25]

    Kilian, N

    M. Kilian, N. J. Mitra, and H. Pottmann. Geometric modeling in shape space.ACM Trans. Graph., 26(3):64–es, July 2007

    2007

  26. [26]

    Klassen, A

    E. Klassen, A. Srivastava, W. Mio, and S. H. Joshi. Analysis of planar shapes using geodesic paths on shape spaces.IEEE Trans. Pattern Anal. Mach. Intell., 26(3):372–383, 2004

    2004

  27. [27]

    Z. Lai, Y . Zhao, H. Liu, Z. Zhao, Q. Lin, H. Shi, X. Yang, M. Yang, S. Yang, Y . Feng, S. Zhang, X. Huang, D. Luo, F. Yang, F. Yang, L. Wang, S. Liu, Y . Tang, Y . Cai, Z. He, T. Liu, Y . Liu, J. Jiang, Linus, J. Huang, and C. Guo. Hunyuan3d 2.5: Towards high-fidelity 3d assets generation with ultimate details.CoRR, abs/2506.16504, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  28. [28]

    X. Liu, C. Gong, and Q. Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net, 2023

    2023

  29. [29]

    Loper, N

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black. SMPL: A skinned multi-person linear model.ACM Transactions on Graphics, 34(6), 2015

    2015

  30. [30]

    J. Lu, J. Lin, H. Dou, A. Zeng, Y . Deng, X. Liu, Z. Cai, L. Yang, Y . Zhang, H. Wang, et al. Dposer-x: Diffusion model as robust 3d whole-body human pose prior. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 9988–9997, 2025

    2025

  31. [31]

    Maesumi, P

    A. Maesumi, P. Guerrero, N. Aigerman, V . Kim, M. Fisher, S. Chaudhuri, and D. Ritchie. Explorable mesh deformation subspaces from unstructured 3d generative models. InSIGGRAPH Asia 2023 Conference Papers, SA ’23, New York, NY , USA, 2023. Association for Computing Machinery

    2023

  32. [32]

    P. W. Michor and D. B. Mumford. Riemannian geometries on spaces of plane curves.Journal of the European Mathematical Society, 8(1):1–48, 2006. 11

    2006

  33. [33]

    Muralikrishnan, S

    S. Muralikrishnan, S. Chaudhuri, N. Aigerman, V . G. Kim, M. Fisher, and N. J. Mitra. GLASS: geometric latent augmentation for shape spaces. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 470–479. IEEE, 2022

    2022

  34. [34]

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas. Pointnet++: deep hierarchical feature learning on point sets in a metric space. InProceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 5105–5114, Red Hook, NY , USA, 2017. Curran Associates Inc

    2017

  35. [35]

    Ranjan, T

    A. Ranjan, T. Bolkart, S. Sanyal, and M. J. Black. Generating 3d faces using convolutional mesh autoencoders. In V . Ferrari, M. Hebert, C. Sminchisescu, and Y . Weiss, editors,Computer Vision - ECCV 2018 - 15th European Conference, Munich, Germany, September 8-14, 2018, Proceedings, Part III, volume 11207 ofLecture Notes in Computer Science, pages 725–74...

    2018

  36. [36]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High-resolution image synthesis with latent diffusion models. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 10674– 10685. IEEE, 2022

    2022

  37. [37]

    Song and S

    Y . Song and S. Ermon. Generative modeling by estimating gradients of the data distribution. In H. M. Wallach, H. Larochelle, A. Beygelzimer, F. d’Alché-Buc, E. B. Fox, and R. Garnett, editors,Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8-14, 2019, Vancouver,...

    2019

  38. [38]

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole. Score-based gen- erative modeling through stochastic differential equations. In9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021

    2021

  39. [39]

    Srivastava, E

    A. Srivastava, E. Klassen, S. H. Joshi, and I. H. Jermyn. Shape analysis of elastic curves in euclidean spaces.IEEE Trans. Pattern Anal. Mach. Intell., 33(7):1415–1428, 2011

    2011

  40. [40]

    Stathopoulos, L

    A. Stathopoulos, L. Han, and D. N. Metaxas. Score-guided diffusion for 3d human recovery. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024, pages 906–915. IEEE, 2024

    2024

  41. [41]

    Q. Tan, L. Gao, Y .-K. Lai, and S. Xia. Variational autoencoders for deforming 3d mesh models. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5841–5850, June 2018

    2018

  42. [42]

    Tevet, S

    G. Tevet, S. Raab, B. Gordon, Y . Shafir, D. Cohen-Or, and A. H. Bermano. Human motion diffusion model. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net, 2023

    2023

  43. [43]

    Vahdat, K

    A. Vahdat, K. Kreis, and J. Kautz. Score-based generative modeling in latent space. In M. Ranzato, A. Beygelzimer, Y . N. Dauphin, P. Liang, and J. W. Vaughan, editors,Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, pages 11287–11302, 2021

    2021

  44. [44]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is all you need. InProceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY , USA,

  45. [45]

    Curran Associates Inc

  46. [46]

    J. Wu, C. Zhang, T. Xue, B. Freeman, and J. Tenenbaum. Learning a probabilistic latent space of object shapes via 3d generative-adversarial modeling. In D. D. Lee, M. Sugiyama, U. von Luxburg, I. Guyon, and R. Garnett, editors,Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016, December 5-...

    2016

  47. [47]

    Xiang, Z

    J. Xiang, Z. Lv, S. Xu, Y . Deng, R. Wang, B. Zhang, D. Chen, X. Tong, and J. Yang. Structured 3d latents for scalable and versatile 3d generation. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025, pages 21469–21480. Computer Vision Foundation / IEEE, 2025

    2025

  48. [48]

    Xiang, Z

    J. Xiang, Z. Lv, S. Xu, Y . Deng, R. Wang, B. Zhang, D. Chen, X. Tong, and J. Yang. Structured 3d latents for scalable and versatile 3d generation. InProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 21469–21480, June 2025

    2025

  49. [49]

    H. Yang, X. Huang, B. Sun, C. L. Bajaj, and Q. Huang. Gencorres: Consistent shape matching via coupled implicit-explicit shape generative models. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024

    2024

  50. [50]

    H. Yang, B. Sun, L. Chen, A. Pavel, and Q. Huang. Geolatent: A geometric approach to latent space design for deformable shape generators.ACM Trans. Graph., 42(6), Dec. 2023

    2023

  51. [51]

    L. Yang, Z. Zhang, Y . Song, S. Hong, R. Xu, Y . Zhao, W. Zhang, B. Cui, and M.-H. Yang. Diffusion models: A comprehensive survey of methods and applications.ACM Comput. Surv., 56(4), Nov. 2023

    2023

  52. [52]

    B. Yi, V . Ye, M. Zheng, L. Müller, G. Pavlakos, Y . Ma, J. Malik, and A. Kanazawa. Estimating body and hand motion in an ego-sensed world.CoRR, abs/2410.03665, 2024

    work page arXiv 2024

  53. [53]

    Zhang, J

    B. Zhang, J. Tang, M. Nießner, and P. Wonka. 3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models.ACM Trans. Graph., 42(4), July 2023

    2023

  54. [54]

    Zhang, B

    S. Zhang, B. L. Bhatnagar, Y . Xu, A. Winkler, P. Kadlecek, S. Tang, and F. Bogo. Rohm: Robust human motion reconstruction via diffusion. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024, pages 14606–14617. IEEE, 2024

    2024

  55. [55]

    K. Zhou, B. L. Bhatnagar, and G. Pons-Moll. Unsupervised shape and pose disentanglement for 3d meshes. In A. Vedaldi, H. Bischof, T. Brox, and J.-M. Frahm, editors,ECCV (22), volume 12367 ofLecture Notes in Computer Science, pages 341–357. Springer, 2020

    2020

  56. [56]

    Y . Zhou, C. Wu, Z. Li, C. Cao, Y . Ye, J. M. Saragih, H. Li, and Y . Sheikh. Fully convolutional mesh autoencoder using efficient spatially varying kernels. In H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, editors,Advances in Neural Information Processing Sys- tems 33: Annual Conference on Neural Information Processing Systems 2020, NeurI...

    2020

  57. [57]

    quality filter

    S. Zuffi, A. Kanazawa, D. Jacobs, and M. J. Black. 3D menagerie: Modeling the 3D shape and pose of animals. InIEEE Conf. on Computer Vision and Pattern Recognition (CVPR), July 2017. 13 A Normalized ARAP Reg Our goal is to compute the derivative of rθ(z) =Tr Eθ(z)− 1 2 H θ(z)E θ(z)− 1 2 =Tr H θ(z)E θ(z)−1 where Eθ(z) =J θ(z) T J θ(z), H θ(z) =J θ(z) T H θ...

    2017