Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

FunPhase models each motion clip as a handful of sinusoids and reconstructs motion as a continuous function, cutting reconstruction error by at least 45% versus prior periodic autoencoders.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 17:25 UTC pith:6PQQRLIC

load-bearing objection A real architectural step forward in phase-based motion models, but the abstract's 'uniform at least 45%' claim is contradicted by the paper's own Table 1. the 3 major comments →

arxiv 2512.09423 v2 pith:6PQQRLIC submitted 2025-12-10 cs.CV

FunPhase: A Periodic Functional Autoencoder for Motion Generation via Phase Manifolds

Marco Pegoraro , Evan Atherton , Bruno Roy , Aliasghar Khani , Arianna Rampini This is my paper
classification cs.CV
keywords motion generationphase manifoldperiodic autoencoderfunctional representationlatent diffusionskeleton-agnosticmotion super-resolutionmotion completion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces FunPhase, a periodic functional autoencoder that turns motion into a continuous spatio-temporal function rather than a sequence of frames. Its central claim is that decomposing each latent channel into a single sinusoid—phase, amplitude, frequency, and offset—yields a compact, interpretable phase manifold that supports reconstruction, generation, super-resolution, body completion, and phase-aligned control in one skeleton-agnostic framework. The authors argue that this phase-structured function space acts as a strong inductive bias, reducing reconstruction error substantially compared to prior periodic autoencoders while matching state-of-the-art motion generation quality. A sympathetic reader would care because the method promises to unify previously separate tasks in motion learning and to make phase-based representations compatible with modern generative diffusion models.

Core claim

The paper's core claim is that motion can be encoded as a continuous function of time and skeleton position using a periodic autoencoder whose latent space is parameterized by per-channel sinusoids. Each latent channel carries four scalars—phase shift, amplitude, frequency, and offset—obtained via a differentiable FFT layer, and the decoder evaluates these sinusoids at arbitrary spatio-temporal coordinates through Perceiver-based cross-attention. This functional formulation replaces discrete frame decoding, enabling arbitrary temporal resolution, generalization across skeletons and datasets, and probabilistic generation via latent diffusion performed directly on the phase parameters. The aut

What carries the argument

The load-bearing mechanism is the channel-wise sinusoidal parameterization of the latent space: every latent channel is modeled as a_c * sin(2π(f_c * T − s_c)) + b_c, with parameters extracted by a differentiable FFT and a learned phase regressor. This single-frequency-per-channel representation enforces periodicity and creates a smooth phase manifold, which the decoder queries as a continuous function. The accompanying domain transformations (phase-to-Cartesian, probit on frequency) make these parameters suitable for Gaussian diffusion, while the Perceiver encoder/decoder pairs handle variable skeleton topology via graph positional encodings.

Load-bearing premise

The claim rests on assuming that every one-second motion window can be faithfully represented as a fixed sum of independent single-frequency sinusoids, one per latent channel, with constant amplitude, frequency, phase, and offset throughout the window.

What would settle it

Take a real 1-second motion clip that contains a clear gait transition, such as a walk-to-run change within the window, encode it with FunPhase, and measure the reconstruction error at the transition frames versus steady-state frames. If the single-frequency-per-channel model is truly capturing motion structure, the error should spike at the transition, revealing the representation's blind spot for non-stationary content.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Motion reconstruction, prediction, generation, super-resolution, and partial-body completion can share one learned phase manifold, simplifying pipelines that currently require separate models.
  • Because decoding is a continuous function evaluated at query coordinates, the same trained model can output motion at any frame rate or skeleton resolution without retraining.
  • Phase-structured latents stabilize latent diffusion: the paper reports lower FID and better physical plausibility than diffusion without the phase bottleneck, suggesting periodicity is a useful generative prior for motion.
  • The compact 4-parameter-per-channel representation enables skeleton-agnostic training across human and animal datasets, opening the door to cross-species motion transfer and control.

Where Pith is reading between the lines

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

  • If single-sinusoid channels suffice for a one-second window, the strongest implication is that local periodicity is the dominant temporal structure in atomic motion segments; an editorially natural test is whether modeling two or three sinusoids per channel (or allowing frequency drift) would unlock longer, non-stationary windows without losing interpretability.
  • The functional decoder could be coupled with physics-based priors (contact forces, dynamics) more naturally than frame-based decoders, since the output is a differentiable function of time and joint coordinates; this is an extension the paper names as future work.
  • The phase manifold's circular structure might support direct interpolation and blending in latent space (e.g., style transfer by adding phase vectors), similar to how phase manifolds helped motion matching in earlier work; the paper only hints at control via a Mixture-of-Experts controller.
  • A concrete falsifiable extension of the paper's logic: if the sinusoidal bottleneck captures true motion structure, then reconstruction error should degrade sharply when the same clip contains two distinct gaits within one window; testing this would separate the power of the representation from the capacity of the Perceiver decoder.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces FunPhase, a functional periodic autoencoder that represents motion as a continuous spatio-temporal function. Each latent channel is modeled as a single sinusoid with learned amplitude, frequency, phase shift, and offset, following the DeepPhase periodic autoencoder but replacing the frame-based convolutional decoder with a Perceiver-based function decoder that can decode at arbitrary temporal and joint resolutions. The latent parameters are transformed (Cartesian phase, probit frequency, raw offset) and used to train a DiT-based latent diffusion model for class-conditional motion generation. The method is evaluated on reconstruction against DeepPhase and other autoencoders, on generation against MLD, ACMDM, CAMDM, AnyTop, and SinMDM, and on qualitative motion-controller and super-resolution/completion tasks. The paper claims substantial and uniform reconstruction improvements over prior periodic autoencoders and competitive generation quality.

Significance. If the results hold, FunPhase would be a valuable contribution: it unifies reconstruction, generation, super-resolution, and partial-body completion in a single interpretable and skeleton-agnostic phase manifold. The functional formulation, Perceiver-based architecture, and diffusion over phase-transformed parameters are well motivated and clearly described. The empirical evaluation includes multiple baselines, ablations, and downstream tasks. The paper also provides honest limitations, though the core claims need tightening to match the reported numbers.

major comments (3)
  1. [Abstract and Table 1] The abstract states 'uniform improvements of at least 45% across all metrics' over prior periodic autoencoders. Table 1 does not support this. Relative improvements (DeepPhase to FunPhase) are: DOG Orientation 0.54→0.34 (37%), DOG ACL 1.768→1.062 (40%), 100STYLE ACL with 32C 1.504→1.371 (8.8%), and with 256C 1.504→1.378 (8.4%); 100STYLE Orientation with 256C is 0.35→0.20 (43%). Thus at least two metrics fail the 45% threshold for every compared variant. The claim must be corrected to reflect metric-specific or 'most metrics' improvements, or the comparison must be recomputed.
  2. [Section 5.1, Table 1] The text says 'Notably, despite operating in a more compact latent space imposed by the periodic parameterization, FUNPHASE still surpasses the Function Autoencoder in reconstruction accuracy.' This is not uniformly true. Comparing FunPhase-256C with Function AE-256C: Position improves (0.36 vs 0.74), but Orientation is worse (0.20 vs 0.05) and NPSS is worse (0.75 vs 0.59). The claim should be qualified to specific metrics (e.g., position and physical plausibility) or the numbers must be revisited.
  3. [Section 4.1, Eq. (1); Section 5.4] The model assumes each latent channel is a single sinusoid with constant amplitude, frequency, phase shift, and offset over the entire 1-second window. This is a stationarity assumption that cannot represent non-stationary segments such as transitions, speed changes, or contact events within a window. The limitation section only acknowledges loss of detail at very low sampling rates, not this representational restriction. Given the paper's claims of generality across skeletons and datasets, the authors should either supply an analysis of reconstruction error on non-periodic or transitional motions, or explicitly scope the method to approximately periodic locomotion.
minor comments (5)
  1. [Table 1] The dataset name is typeset as '100STYLES' (missing space and extra S). Also 'MLD-V AE' and 'ACMDM-AE' have inconsistent spacing.
  2. [Table 2] The column 'ACCL' is likely meant to be 'ACL' (Average Curve Length) as used in Table 1. Please make consistent.
  3. [Section 7.2 (Supplementary)] The conditioning latent is written 'DF unP hase' in the equation; likely a typo for 'FunPhase'.
  4. [Section 8.3 (Supplementary)] The text refers to 'Table 1 of the main paper' when comparing Function Diffusion and FunPhase; the relevant table is Table 2.
  5. [Section 5.2] The sentence 'this results in clear gains across all metrics' is too strong. In Table 2, FunPhase has slightly worse Coherence and ACCL than Function Diff, and worse Diversity than MLD and CAMDM. Please qualify.

Circularity Check

0 steps flagged

No significant circularity: the derivation is architecturally explicit and benchmarked externally; the 45% claim is a reporting inconsistency, not a circular step.

full rationale

The core derivation is self-contained in the relevant sense. The encoder maps motion to periodic parameters theta_c = [s_c, a_c, f_c, b_c] (Eq. 1); the decoder evaluates those sinusoids at queried coordinates; and the diffusion model operates on an invertible reparameterization of the same parameters (Eq. 9). There is no quantity that is fitted to a target and then renamed as a prediction of that same target. Reconstruction, super-resolution, completion, and generation all use the same explicit encoder/decoder function evaluations and are tested against external baselines: DeepPhase, MLD-VAE, and ACMDM-AE in Table 1; MLD, ACMDM, and CAMDM in Table 2; SinMDM and AnyTop in Table 3, using released checkpoints or official retraining where stated. The periodic-decomposition idea is credited to the external prior work DeepPhase [26] and the function-space decoder design to FunDiff [31]; the authors' own references ([12] UniMoGen, [13] survey) appear only in related-work enumeration and are not load-bearing, and no uniqueness theorem is imported from the authors' own prior work. The abstract's claim of 'uniform improvements of at least 45% across all metrics' is not supported by the paper's own Table 1 (e.g., DOG orientation 0.54 -> 0.34 is roughly 37%, and 100STYLE ACL 1.504 -> 1.371 is roughly 9%), and the phase-transformation ablation in Table 6 shows only a marginal FID change (1.28 vs 1.27, accuracy 34.83 vs 37.6). These are claim-consistency and empirical-support issues, not circular derivation. The single-sinusoid stationarity assumption (Eq. 1) and the acknowledged loss of fine-grained detail at very low sampling rates (Sec. 5.4) are representational limitations, not definitional circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The paper does not introduce new physical entities or exotic mathematical objects; its contributions are architectural (function-space periodic decoding, diffusion on phase parameters). The ledger above captures the hand-chosen hyperparameters and the domain assumptions that carry the method: the sinusoid-per-channel hypothesis, the skeleton-agnostic encoding hypothesis, and the diffusion-compatibility hypothesis. These are the places where a false premise would invalidate the central claims.

free parameters (4)
  • latent_channels_C = 256 (main results; 16/32/128 ablated)
    The number of latent channels controls capacity and is selected by validation performance (Table 4). It directly affects the reconstruction numbers that anchor the central claims.
  • loss_weights = 0.5 (rotation+root), 0.5 (FK), 0.01 (foot)
    Hand-chosen composite loss weights in Eq. 8; no sensitivity analysis is reported, so the balance of reconstruction, FK, and foot-contact terms is a free design choice.
  • min_SNR_gamma = 5
    Min-SNR loss weighting constant from [8], adopted without sweep; affects diffusion training stability and generation quality.
  • diffusion_schedule_and_steps = beta_min=1e-4, beta_max=0.02, T=1000, DDIM S=900
    Linear noise schedule and DDIM sampling steps chosen as standard; no ablation shown for these values, and they influence all generation metrics.
axioms (5)
  • domain assumption Motion within a 1-second window is well represented as a combination of a small number of sinusoids with constant amplitude, frequency, phase shift, and offset (Eq. 1, Sec. 4.1).
    This is the key inductive bias inherited from DeepPhase. Non-stationary transitions (sit-to-stand, style changes, contact events) cannot be represented; the paper's own limitation (Sec. 5.4) concedes detail loss at low sampling rates.
  • domain assumption Graph Laplacian eigenvectors / heat-diffusion probe encodings are sufficient to align latent spaces across heterogeneous skeletons.
    Sec. 4.1 and Supplementary 7.1 rely on these encodings for skeleton-agnostic generalization. Only the ZOO experiment tests this, and it reports aggregate diversity metrics, not per-skeleton retargeting quality.
  • ad hoc to paper The transformed phase parameters (Cartesian phase, probit frequency, raw offset) are sufficiently close to Gaussian for the DiT velocity-prediction diffusion to work.
    Eq. 9 domain transforms are introduced specifically to make Gaussian diffusion applicable. The supporting ablation (Table 6) is run on a 20% subset and shows only a marginal FID change (1.28 to 1.27), so the necessity of the transformation is weakly supported.
  • standard math The continuous 6D rotation representation and forward-kinematics/contact losses are valid and sufficient for physically plausible motion.
    Uses the 6D rotation representation of [38] and standard FK/foot-contact losses (Eqs. 4-8); these are standard tools in the graphics community and not introduced by this paper.
  • domain assumption The pretrained motion classifier from CAMDM [2] yields features and labels that faithfully capture perceptual style quality for FID and Accuracy.
    Table 2's headline FID and Accuracy numbers depend on this classifier; its reliability on the 100STYLE dataset is assumed, not validated in the paper.

pith-pipeline@v1.3.0-alltime-deepseek · 15170 in / 17336 out tokens · 163033 ms · 2026-08-03T17:25:46.536290+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of FunPhase: A Periodic Functional Autoencoder for Motion Generation via Phase Manifolds." pith.science (2026). https://pith.science/paper/6PQQRLIC

@misc{pith2026251209423,
  author       = {Pith},
  title        = {Pith review of: FunPhase: A Periodic Functional Autoencoder for Motion Generation via Phase Manifolds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6PQQRLIC}},
  note         = {Machine review of arXiv:2512.09423}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Learning natural body motion remains challenging due to the strong coupling between spatial geometry and temporal dynamics. Embedding motion in phase manifolds, latent spaces that capture local periodicity, has proven effective for motion prediction; however, existing approaches are tied to fixed skeletons and narrow motion distributions, limiting their applicability across diverse settings. We introduce FunPhase, a functional periodic autoencoder that learns a phase manifold for motion and replaces discrete temporal decoding with a function-space formulation, enabling smooth trajectories that can be sampled at arbitrary temporal resolutions. FunPhase unifies motion prediction and generation within a single interpretable phase manifold, enabling motion generation via latent diffusion, generalizes across skeletons and datasets, and supports downstream tasks such as motion super-resolution and partial-body completion. Our model achieves substantially lower reconstruction error than prior periodic autoencoder baselines, achieving uniform improvements of at least $45\%$ across all metrics, while enabling a broader range of applications and performing on par with state-of-the-art motion generation methods.

Figures

Figures reproduced from arXiv: 2512.09423 by Aliasghar Khani, Arianna Rampini, Bruno Roy, Evan Atherton, Marco Pegoraro.

Figure 1
Figure 1. Figure 1: FunPhase is a functional periodic autoencoder that learns a phase-structured manifold for motion, enabling smooth, continuous [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Periodic Function Autoencoder (FunPhase) architecture. The figure illustrates the separated processing of joint rotations and root positions through Perceiver-based encoder–decoder modules. The latent space is decomposed by a Fast Fourier Transform (FFT) layer in its periodic components (Phase shift, Amplitude, Frequency, Bias) to achieve an even more compact representa￾tion and enforce per… view at source ↗
Figure 3
Figure 3. Figure 3: Phase Manifold. The plots show the phase manifolds obtained with DeepPhase and FunPhase, alongside the original motion features. All encoded sequences correspond to a dog￾running motion. results. Notably, despite operating in a more compact la￾tent space imposed by the periodic parameterization, FUN￾PHASE still surpasses the Function Autoencoder in recon￾struction accuracy. This indicates that introducing … view at source ↗
Figure 5
Figure 5. Figure 5: Diffusion examples on 100STYLE. On the left we show and example of generation from a sparse set of key frames (in green). On the right we show an example of body completion of the right leg (in pink). motion in a continuous functional space. The key differ￾ence lies in our phase-formulation: by decomposing mo￾tion into phase components and applying the transforma￾tions in Equation 9, FUNPHASE introduces te… view at source ↗
Figure 7
Figure 7. Figure 7: Phase Transformation. We plot the distribution of the latent periodic parameterization before and after the phase trans￾formation applied in the latent diffusion model. 8.3. Phase Transformation The periodic parameterization, while compact and inter￾pretable, poses challenges for diffusion modeling due to the domain and distribution of its parameters. To address this, we apply domain transformations that m… view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. PHASOR: Phase-Anchored Universal Action Representations for Humanoid Embodiments

    cs.RO 2026-06 unverdicted novelty 6.0

    PHASOR factorizes motion into an FFT-based phase manifold and pose branch with semantic distillation to produce a cross-embodiment, human-anchored action embedding space for humanoid robots.

Reference graph

Works this paper leans on

42 extracted references · 10 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Motion signal pro- cessing

    Armin Bruderlin and Lance Williams. Motion signal pro- cessing. InProceedings of the 22nd annual conference on Computer graphics and interactive techniques, pages 97– 104, 1995. 2

    1995

  2. [2]

    Taming diffusion probabilistic mod- els for character control

    Rui Chen, Mingyi Shi, Shaoli Huang, Ping Tan, Taku Ko- mura, and Xuelin Chen. Taming diffusion probabilistic mod- els for character control. InACM SIGGRAPH 2024 Confer- ence Papers, pages 1–10, 2024. 2, 6, 7

    2024

  3. [3]

    Executing your commands via motion diffusion in latent space

    Xin Chen, Biao Jiang, Wen Liu, Zilong Huang, Bin Fu, Tao Chen, and Gang Yu. Executing your commands via motion diffusion in latent space. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18000–18010, 2023. 2, 6, 7

    2023

  4. [4]

    Go to zero: Towards zero-shot motion generation with million-scale data.arXiv preprint arXiv:2507.07095,

    Ke Fan, Shunlin Lu, Minyue Dai, Runyi Yu, Lixing Xiao, Zhiyang Dou, Junting Dong, Lizhuang Ma, and Jingbo Wang. Go to zero: Towards zero-shot motion generation with million-scale data.arXiv preprint arXiv:2507.07095,

    Pith/arXiv arXiv

  5. [5]

    Continuous-time functional diffusion processes.Advances in Neural Information Processing Systems, 36:37370–37400,

    Giulio Franzese, Giulio Corallo, Simone Rossi, Markus Heinonen, Maurizio Filippone, and Pietro Michiardi. Continuous-time functional diffusion processes.Advances in Neural Information Processing Systems, 36:37370–37400,

  6. [6]

    Anytop: Character ani- mation diffusion with any topology

    Inbar Gat, Sigal Raab, Guy Tevet, Yuval Reshef, Amit Haim Bermano, and Daniel Cohen-Or. Anytop: Character ani- mation diffusion with any topology. InProceedings of the Special Interest Group on Computer Graphics and Interac- tive Techniques Conference Conference Papers, pages 1–10,

  7. [7]

    A neural temporal model for hu- man motion prediction

    Anand Gopalakrishnan, Ankur Mali, Dan Kifer, Lee Giles, and Alexander G Ororbia. A neural temporal model for hu- man motion prediction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12116–12125, 2019. 6, 2

    2019

  8. [8]

    Efficient diffu- sion training via min-snr weighting strategy

    Tiankai Hang, Shuyang Gu, Chen Li, Jianmin Bao, Dong Chen, Han Hu, Xin Geng, and Baining Guo. Efficient diffu- sion training via min-snr weighting strategy. InProceedings of the IEEE/CVF international conference on computer vi- sion, pages 7441–7451, 2023. 2

    2023

  9. [9]

    Phase- functioned neural networks for character control.ACM Transactions on Graphics (TOG), 36(4):1–13, 2017

    Daniel Holden, Taku Komura, and Jun Saito. Phase- functioned neural networks for character control.ACM Transactions on Graphics (TOG), 36(4):1–13, 2017. 2, 3

    2017

  10. [10]

    Perceiver: General perception with iterative attention

    Andrew Jaegle, Felix Gimeno, Andy Brock, Oriol Vinyals, Andrew Zisserman, and Joao Carreira. Perceiver: General perception with iterative attention. InInternational confer- ence on machine learning, pages 4651–4664. PMLR, 2021. 4

    2021

  11. [11]

    Motiongpt: Human motion as a foreign lan- guage.Advances in Neural Information Processing Systems, 36:20067–20079, 2023

    Biao Jiang, Xin Chen, Wen Liu, Jingyi Yu, Gang Yu, and Tao Chen. Motiongpt: Human motion as a foreign lan- guage.Advances in Neural Information Processing Systems, 36:20067–20079, 2023. 2

    2023

  12. [12]

    Unimogen: Universal motion generation.arXiv preprint arXiv:2505.21837, 2025

    Aliasghar Khani, Arianna Rampini, Evan Atherton, and Bruno Roy. Unimogen: Universal motion generation.arXiv preprint arXiv:2505.21837, 2025. 2

    Pith/arXiv arXiv 2025

  13. [13]

    Motion generation: A survey of generative approaches and benchmarks.arXiv preprint arXiv:2507.05419, 2025

    Aliasghar Khani, Arianna Rampini, Bruno Roy, Larasika Nadela, Noa Kaplan, Evan Atherton, Derek Cheung, and Jacky Bibliowicz. Motion generation: A survey of generative approaches and benchmarks.arXiv preprint arXiv:2507.05419, 2025. 1, 3

    Pith/arXiv arXiv 2025

  14. [14]

    Scalable group choreog- raphy via variational phase manifold learning

    Nhat Le, Khoa Do, Xuan Bui, Tuong Do, Erman Tjiputra, Quang D Tran, and Anh Nguyen. Scalable group choreog- raphy via variational phase manifold learning. InEuropean Conference on Computer Vision, pages 293–311. Springer,

  15. [15]

    Walkthedog: Cross-morphology motion alignment via phase manifolds

    Peizhuo Li, Sebastian Starke, Yuting Ye, and Olga Sorkine- Hornung. Walkthedog: Cross-morphology motion alignment via phase manifolds. InACM SIGGRAPH 2024 Conference Papers, pages 1–10, 2024. 2

    2024

  16. [16]

    Hier- archical spacetime control

    Zicheng Liu, Steven J Gortler, and Michael F Cohen. Hier- archical spacetime control. InProceedings of the 21st an- nual conference on Computer graphics and interactive tech- niques, pages 35–42, 1994. 2

    1994

  17. [17]

    Ian Mason, Sebastian Starke, and Taku Komura. Real-time style modelling of human locomotion via feature-wise trans- formations and local motion phases.Proceedings of the ACM on Computer Graphics and Interactive Techniques, 5(1):1– 18, 2022. 3, 6, 1

    2022

  18. [18]

    Absolute coordinates make motion generation easy.arXiv preprint arXiv:2505.19377, 2025

    Zichong Meng, Zeyu Han, Xiaogang Peng, Yiming Xie, and Huaizu Jiang. Absolute coordinates make motion generation easy.arXiv preprint arXiv:2505.19377, 2025. 2, 6, 7, 3

    Pith/arXiv arXiv 2025

  19. [19]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 4195–4205,

  20. [20]

    Single motion diffusion

    Sigal Raab, Inbal Leibovitch, Guy Tevet, Moab Arar, Amit H Bermano, and Daniel Cohen-Or. Single motion diffusion. arXiv preprint arXiv:2302.05905, 2023. 2, 6, 8

    Pith/arXiv arXiv 2023

  21. [21]

    Progressive distillation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512, 2022

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512, 2022. 2

    Pith/arXiv arXiv 2022

  22. [22]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 2

    Pith/arXiv arXiv 2010

  23. [23]

    Motion in-betweening with phase manifolds.Pro- ceedings of the ACM on Computer Graphics and Interactive Techniques, 6(3):1–17, 2023

    Paul Starke, Sebastian Starke, Taku Komura, and Frank Steinicke. Motion in-betweening with phase manifolds.Pro- ceedings of the ACM on Computer Graphics and Interactive Techniques, 6(3):1–17, 2023. 2

    2023

  24. [24]

    Neural state machine for character-scene interactions.ACM Transactions on Graphics, 38(6):178, 2019

    Sebastian Starke, He Zhang, Taku Komura, and Jun Saito. Neural state machine for character-scene interactions.ACM Transactions on Graphics, 38(6):178, 2019. 2

    2019

  25. [25]

    Local motion phases for learning multi-contact char- acter movements.ACM Transactions on Graphics, 39(4),

    Sebastian Starke, Yiwei Zhao, Taku Komura, and Kazi Za- man. Local motion phases for learning multi-contact char- acter movements.ACM Transactions on Graphics, 39(4),

  26. [26]

    Deepphase: Periodic autoencoders for learning motion phase manifolds

    Sebastian Starke, Ian Mason, and Taku Komura. Deepphase: Periodic autoencoders for learning motion phase manifolds. ACM Transactions on Graphics (ToG), 41(4):1–13, 2022. 2, 3, 4, 6, 7, 8 9

    2022

  27. [27]

    Human motion dif- fusion model.arXiv preprint arXiv:2209.14916, 2022

    Guy Tevet, Sigal Raab, Brian Gordon, Yonatan Shafir, Daniel Cohen-Or, and Amit H Bermano. Human motion dif- fusion model.arXiv preprint arXiv:2209.14916, 2022. 2

    Pith/arXiv arXiv 2022

  28. [28]

    Truebones motion-capture dataset.https: //truebones.gumroad.com/l/skZMC, 2025

    TrueBones. Truebones motion-capture dataset.https: //truebones.gumroad.com/l/skZMC, 2025. Ac- cessed: 2025-07-02. 6, 3

    2025

  29. [29]

    Dif- fusionphase: Motion diffusion in frequency domain.arXiv preprint arXiv:2312.04036, 2023

    Weilin Wan, Yiming Huang, Shutong Wu, Taku Komura, Wenping Wang, Dinesh Jayaraman, and Lingjie Liu. Dif- fusionphase: Motion diffusion in frequency domain.arXiv preprint arXiv:2312.04036, 2023. 2

    Pith/arXiv arXiv 2023

  30. [30]

    Spatio-temporal manifold learning for human motions via long-horizon modeling.IEEE transactions on visualiza- tion and computer graphics, 27(1):216–227, 2019

    He Wang, Edmond SL Ho, Hubert PH Shum, and Zhanxing Zhu. Spatio-temporal manifold learning for human motions via long-horizon modeling.IEEE transactions on visualiza- tion and computer graphics, 27(1):216–227, 2019. 2

    2019

  31. [31]

    Fundiff: Diffusion models over function spaces for physics-informed generative modeling.arXiv preprint arXiv:2506.07902,

    Sifan Wang, Zehao Dou, Tong-Rui Liu, and Lu Lu. Fundiff: Diffusion models over function spaces for physics-informed generative modeling.arXiv preprint arXiv:2506.07902,

    arXiv

  32. [32]

    Animo: Species-aware model for text-driven animal motion generation

    Xuan Wang, Kai Ruan, Xing Zhang, and Gaoang Wang. Animo: Species-aware model for text-driven animal motion generation. InProceedings of the Computer Vision and Pat- tern Recognition Conference, pages 1929–1939, 2025. 2

    1929

  33. [33]

    Swallowing the bit- ter pill: Simplified scalable conformer generation.arXiv preprint arXiv:2311.17932, 2023

    Yuyang Wang, Ahmed A Elhag, Navdeep Jaitly, Joshua M Susskind, and Miguel Angel Bautista. Swallowing the bit- ter pill: Simplified scalable conformer generation.arXiv preprint arXiv:2311.17932, 2023. 3

    Pith/arXiv arXiv 2023

  34. [34]

    Scaling large motion mod- els with million-level human motions

    Ye Wang, Sipeng Zheng, Bin Cao, Qianshan Wei, Weishuai Zeng, Qin Jin, and Zongqing Lu. Scaling large motion mod- els with million-level human motions. InForty-second Inter- national Conference on Machine Learning, 2025. 2

    2025

  35. [35]

    Physdiff: Physics-guided human motion diffusion model

    Ye Yuan, Jiaming Song, Umar Iqbal, Arash Vahdat, and Jan Kautz. Physdiff: Physics-guided human motion diffusion model. 2023 ieee. InCVF International Conference on Com- puter Vision (ICCV), pages 15964–15975, 2022. 2

    2023

  36. [36]

    Mode-adaptive neural networks for quadruped motion con- trol.ACM Transactions on Graphics (ToG), 37(4):1–11,

    He Zhang, Sebastian Starke, Taku Komura, and Jun Saito. Mode-adaptive neural networks for quadruped motion con- trol.ACM Transactions on Graphics (ToG), 37(4):1–11,

  37. [37]

    Im- proved techniques for maximum likelihood estimation for diffusion odes

    Kaiwen Zheng, Cheng Lu, Jianfei Chen, and Jun Zhu. Im- proved techniques for maximum likelihood estimation for diffusion odes. InInternational Conference on Machine Learning. PMLR, 2023. 5

    2023

  38. [38]

    On the con- tinuity of rotation representations in neural networks

    Yi Zhou, C Barnes, J Lu, J Yang, and H Li. On the con- tinuity of rotation representations in neural networks. 2019 ieee. InCVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018. 4

    2019

  39. [39]

    Human motion generation: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(4):2430– 2449, 2023

    Wentao Zhu, Xiaoxuan Ma, Dongwoo Ro, Hai Ci, Jinlu Zhang, Jiaxin Shi, Feng Gao, Qi Tian, and Yizhou Wang. Human motion generation: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(4):2430– 2449, 2023. 2, 3, 6 10 FunPhase: A Periodic Functional Autoencoder for Motion Generation via Phase Manifolds Supplementary Material

    2023

  40. [40]

    The code will be released upon publication

    Implementation Details This section provides additional implementation details and the experimental configurations. The code will be released upon publication. 7.1. FunPhase Positional Encoding.To provide the model with informa- tion about temporal structure and skeletal topology, we em- ploy two types of positional encodings: (i) Temporal Encoding:We use...

  41. [41]

    Ablations In this section, we show the ablation studies that guided our final model design. 8.1. FunPhase We trained the ablation models using20%of the 100STYLE training set for60epochs. Model size.We first ablate the number of latent chan- nels in FunPhase. Table 4 reports the reconstruction errors for models with32,128, and256channels. As expected, the ...

  42. [42]

    In Table 7, we evaluate the reconstruction performance of our model when conditioned on keyframes sampled at in- creasing temporal intervals

    Reconstruction error under increasing keyframe distances. In Table 7, we evaluate the reconstruction performance of our model when conditioned on keyframes sampled at in- creasing temporal intervals. We compare Function AE and FunPhase against a SLERP baseline, which linearly in- terpolates the root trajectory and joint rotations between keyframes. FunPha...