Pith. sign in

REVIEW 2 major objections 3 minor 53 references

MUGEN claims that a single continuous latent-slot representation, written and read by one language model, lets a unified motion-generation-and-understanding system beat discrete-codebook pipelines on retrieval and alignment while decoding e

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 ↗

MUGEN lets a language model generate and read motion through a few continuous latent slots, achieving competitive retrieval and captioning with one draw and K language-model steps.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection MUGEN is a well-executed systems paper with a real architectural contribution, but the unnamed perceptual-loss encoder means the headline retrieval/FID numbers need a re-audit before the central claim is fully trusted. the 2 major comments →

arxiv 2607.27581 v1 pith:AP5F6KM4 submitted 2026-07-30 cs.LG

MUGEN: A Unified Framework for Efficient Motion Understanding and Generation

Zhankai Ye , Yukai Jin , Bingyang Wei , Bofan Li , Yusen Wu , Fangyi Li , Shangqian Gao , Xin Liu This is my paper
classification cs.LG
keywords motion understandingtext-to-motion generationcontinuous latent slotsadaptive-length autoencoderdepth routinglow-rank factor headunified motion-language modelsingle-draw sampling
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 reading

The paper sets out to prove that a unified motion–language system does not need a discrete motion codebook or iterative decoding to be competitive. Its thesis is that one compact continuous representation — K latent slots from an adaptive-length autoencoder — can serve simultaneously as the target the language model generates from text and as the input the same language model reads back for captioning and retrieval. If correct, the expensive machinery of residual codebooks, masked refinement, and denoising chains is not a necessary price for quality. MUGEN reports leading retrieval and alignment scores at a decoding cost of K language-model steps, one draw, and one decoder pass, with the one reported weakness being FID, which the authors pin on the single-draw sampler rather than on the continuous representation via an oracle residual experiment.

Core claim

The paper's central claim is that a small set of continuous latent slots produced by an adaptive-length autoencoder can serve as the sole motion interface for both generation and understanding, once the language model reads each slot from a prompt-conditioned mixture of transformer depths and draws the whole latent set from a calibrated low-rank-plus-diagonal Gaussian. On this interface, the authors report that a K-step autoregressive rollout, one draw, and one decoder pass suffice to lead language-model baselines on FID on HumanML3D, to achieve the best CIDEr and BLEU@4 captioning scores, and to surpass the discrete-token state of the art on every retrieval and alignment metric on SnapMoGen

What carries the argument

The Adaptive-Length AutoEncoder (ALAE) compresses any-length motion into exactly K continuous latent slots via learned latent queries cross-attending to a temporal memory, then decodes frames from those slots using sinusoidal phase queries; it is the system's only motion representation. Depth-routed hidden states let each slot read from a prompt-conditioned weighted mixture over transformer layers, so slots can specialize (one reads deep, another scans mid-depth) instead of all reading the final layer. The calibrated low-rank factor head predicts a low-rank-plus-diagonal Gaussian over the flattened K·D latent, so a single draw carries text-conditional, cross-slot variation; a decoupled ancho

Load-bearing premise

The load-bearing premise is that the frozen 'dataset-specific motion encoder' whose perceptual loss trains the ALAE is not effectively the same feature space as the official benchmark evaluator that computes R-Precision and FID; if they coincide, the headline retrieval and fidelity numbers are partly a training objective rather than an independent measure of the one-draw continuous interface.

What would settle it

Run the full pipeline twice: once with the ALAE perceptual loss trained against a motion encoder that provably cannot see the benchmark's evaluator features (e.g., trained on held-out joints or a different dataset), and once without the perceptual loss entirely. If MUGEN's R@1-above-real-motion and FID advantage survives both runs, the central claim stands; if either collapses, the reported advantage is attributable to the frozen encoder/evaluator overlap rather than to the continuous-slot one-draw design.

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

If this is right

  • A unified motion–language system can share one continuous representation without codebook hierarchies or iterative decoding; joint training improves generation FID over a generation-only twin.
  • Text-to-motion decoding costs K language-model steps, one draw, and one decoder pass, yielding about 9 ms per motion on HumanML3D and 6–14× lower latency than the strongest masked-codebook and unified baselines.
  • On HumanML3D, retrieval precision and matching distance can exceed the real-motion reference under the standard evaluator; on SnapMoGen, every retrieval rank and CLIP score beat the discrete-token state of the art.
  • The residual FID gap to masked-codebook pipelines is attributed to the single-draw sampler rather than the continuous representation, since decoding with half of the true latent residual brings FID to 6.89 while batch-shuffled residuals collapse R@1.
  • The understanding branch needs no captioning-specific training: the same checkpoint that generates motion leads retrieval ranks, BLEU@4, and CIDEr on HumanML3D.

Where Pith is reading between the lines

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

  • Editorial inference: if the FID gap truly lives in the sampler, a single-step conditional sampler (e.g., a consistency or flow-matching head over the same slots) could close that gap while keeping the understanding branch intact — but that would likely abandon the literal 'one draw' budget unless distilled to a single step.
  • Editorial inference: the retrieval-above-real-motion result on HumanML3D is suspicious enough that the clean test is to re-run the pipeline with an ALAE perceptual encoder trained without access to the benchmark's evaluator features; if the margin shrinks, the shared interface is less responsible than the paper claims.
  • Editorial inference: since captioning quality did not improve with billion-scale language backbones, further motion-to-text gains likely require a richer motion representation (more slots, hierarchical slots, or slot semantics), not larger language models.
  • Editorial inference: the K-scaling result — retrieval peaks at K=2–4 and falls beyond it — suggests the optimal slot budget tracks text–motion correspondence rather than reconstruction fidelity; a testable extension is to let K grow with caption complexity or verb event structure.
Share X Bluesky LinkedIn Reddit HN

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

2 major / 3 minor

Summary. The paper proposes MUGEN, a unified motion–language framework in which a single continuous latent interface — K slots produced by an adaptive-length autoencoder (ALAE) — serves both text-to-motion generation and motion-to-text understanding. A GPT-2 backbone autoregressively predicts the latent slots in K steps; a depth router lets each slot read from a learned mixture of transformer layers, and a calibrated low-rank factor head draws the whole latent set in one structured Gaussian sample. The understanding branch reads the same frozen slots through a projector into the language model. On HumanML3D and SnapMoGen the paper reports strong retrieval, captioning, and FID numbers, with a claimed decoding cost of 9 ms per motion, sampling temperature and checkpoints chosen on validation only, and test metrics computed with the official benchmark evaluators. The central claim is that discrete motion tokenization and iterative decoding are not required for a competitive unified motion–language system.

Significance. If the results hold, MUGEN is a significant contribution: it demonstrates that a compact continuous representation can support both directions of motion–language modeling with large inference savings, and it provides an unusually transparent evaluation protocol (validation-only temperature/checkpoint selection, official 20-replication protocol, release of code and checkpoints). The depth-routing and calibrated low-rank factor head are well-motivated, and the oracle-residual and unified-training ablations are informative. However, the paper never identifies the "frozen dataset-specific motion encoder" used for the ALAE perceptual loss, which is weighted λperc=10 and is also used in the joint training loss. Because Appendix B defines all reported metrics in the feature space of the official benchmark evaluators, this omission is load-bearing for the headline retrieval and FID claims. The paper itself concedes that part of the retrieval effect belongs to the evaluator, but it does not address the training-side analogue. With the encoder identified and an independent-evaluator check, the contribution would be solid; without it, the key comparisons are not fully verifiable.

major comments (2)
  1. [Sec. 3.1; App. B; Table 7] The ALAE objective includes "a perceptual loss from a frozen dataset-specific motion encoder" (Sec. 3.1), weighted λperc=10 (Table 7), and the same perceptual terms appear in the joint objective governing generation (Sec. 3.3). The encoder is never named. Appendix B defines every reported metric in the feature space of the official benchmark evaluators (Guo et al. 2022a for HumanML3D; Guo et al. 2025 for SnapMoGen). If the perceptual encoder is, or was trained with, those evaluators, then the latent slots and the generated motions are explicitly optimized to preserve the exact features used to compute FID, R-Precision, MM-Dist, and CLIP. The retrieval above the real-motion reference (Tables 1 and 3) and the competitive FID would then be partly a training objective, and the comparison with baselines that did not train on those features would not be apples-to-apples. The concession in Sec.
  2. [App. H.2; Sec. 4.2] The oracle-residual diagnosis attributes the residual FID gap to the single-draw sampler by decoding z = μ + s(z*−μ) and observing FID drop to 6.89 at s=0.5. But z* is obtained from the same ALAE encoder, and if that encoder's perceptual training objective is in the evaluator's feature space, the oracle experiment is conducted entirely in a space the system was trained to match. The experiment therefore does not distinguish "the sampler fails to recover variance the representation contains" from "the representation is overfit to the evaluator and the sampler cannot reproduce that overfit." To support the attribution, the oracle-residual test should be repeated with an independent evaluator or in raw motion space, and the sampled FID should be reported under an independent evaluator as well.
minor comments (3)
  1. [Tables 1–3] The main tables quote baseline numbers from original papers or benchmark reports rather than re-running them; state this explicitly in every table caption and clarify whether any baselines were re-run. HumanML3D rows lack confidence intervals while SnapMoGen rows have them; state which rows carry intervals and why.
  2. [Fig. 4; Sec. 4.5] The text says k0 keeps 98% of its routing mass in layers 9–11, while the figure caption says k0 stays fixed on layer 10. Please reconcile these statements.
  3. [App. H.8] The backbone ablation is careful to include a bf16 GPT-2 control, but all billion-scale rows are single-seed runs. State in the main text that the backbone comparison is single-seed and that FID differences of a few thousandths are not treated as significant.

Circularity Check

0 steps flagged

No demonstrated circularity: the unresolved perceptual-encoder identity is a transparency concern, not a reduction.

full rationale

The paper's empirical chain is self-contained with respect to the benchmark protocol: ALAE is trained once, then frozen; the GPT-2 backbone is trained jointly for generation and understanding; and all reported metrics are computed on held-out test splits under each benchmark's official 20-replication protocol, with temperature and checkpoint selection restricted to the validation split (App. A, App. G). I find no step in which a reported result is equivalent by construction to an input. The latent-budget K ablation is reported on the test split (Tables 5, 8) and the paper chooses K from it, but K is a hyperparameter sweep rather than a fitted quantity, so this is a methodological caveat about selection, not a circular reduction. The one notable transparency gap is the 'frozen dataset-specific motion encoder' used for the ALAE perceptual loss (Sec. 3.1). Appendix B defines every metric in the feature space of each benchmark's official evaluator, and the paper never identifies the perceptual encoder nor states explicitly that it differs from that evaluator. If the two encoders were the same, the headline FID and R-Precision numbers would be partly a training objective rather than an independent measure; but the paper provides no equation or statement exhibiting that identity, so per the requirement to demonstrate circularity rather than speculate, this remains an unresolved correctness/transparency risk, not a demonstrated circular step. The paper itself concedes in Sec. 4.2 that 'part of the effect belongs to the evaluator' for retrieval, which is a caveat rather than a circular derivation. The only same-author citation (GeoMotionGPT, Ye et al. 2026) appears in related work and is not load-bearing for any claim. Overall, the central derivation—continuous slots shared between generation and understanding, evaluated on external benchmarks—does not reduce to its inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claims rest on standard benchmark assumptions and several hand-chosen hyperparameters. The main unverified premise is the identity of the frozen perceptual encoder used to train ALAE: if it overlaps the official evaluator, the headline metrics are partly a training objective. No new physical entities or mediators are introduced; the latent slots, router, and factor head are learned parameters and architectural components.

free parameters (5)
  • Latent budget K = 2 (HumanML3D), 4 (SnapMoGen)
    Sets the number of continuous slots and inference steps; K-scaling sweep selected the smallest budget that holds quality, central to the efficiency claim.
  • Sampling temperature tau = 0.6 on both benchmarks at the reported operating point
    Validation-only sweep over [0.5, 1.0]; directly controls the FID versus R@1 trade-off; calibrated sampling produces the reported FID.
  • Factor rank r = 64
    Rank of the low-rank covariance head; chosen by hand and fixed across all K and both datasets.
  • Router scales s_s, s_d = 4, 4
    Caps the router logit margin; set by hand and kept across experiments.
  • Loss weights (lambda_lat, lambda_m2t, lambda_MI, lambda_perc) = 2, 1, 0.05, 10
    Balances the joint objective; lambda_m2t is ablated but not tuned per dataset; perceptron weight is large and tied to the unnamed perceptual encoder.
axioms (5)
  • domain assumption Official benchmark evaluators (HumanML3D contrastive evaluator, SnapMoGen TMR-style dual encoder) are accepted as ground truth for quality.
    All headline metrics are computed in these feature spaces (App. B); the paper itself concedes the real-motion reference is crossed by many baselines, so evaluator behavior is load-bearing.
  • domain assumption ALAE's perceptual loss uses a 'frozen dataset-specific motion encoder' that is not named and is assumed independent of the official evaluator.
    Sec. 3.1; if this encoder is the evaluator's motion encoder, the autoencoder is trained on the same features used to score it.
  • domain assumption A low-rank-plus-diagonal Gaussian over the K x D latent set is expressive enough for text-conditioned variation.
    Sec. 3.2 and App. H.3; supported by ablations, but still a modeling assumption about the latent family.
  • domain assumption GPT-2 with 12 layers is an adequate backbone; larger LMs do not improve the joint objective.
    App. H.8 is a single-seed ablation and the authors caution against reading small FID gaps; this is empirical evidence, not a theorem.
  • domain assumption The frozen ALAE decoder continues to decode well after the LM learns to predict slots.
    Decoder is never updated in joint training; the oracle residual test (App. H.2) supports headroom, but decoder quality still bounds generation.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MUGEN: A Unified Framework for Efficient Motion Understanding and Generation." pith.science (2026). https://pith.science/paper/AP5F6KM4

@misc{pith2026260727581,
  author       = {Pith},
  title        = {Pith review of: MUGEN: A Unified Framework for Efficient Motion Understanding and Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AP5F6KM4}},
  note         = {Machine review of arXiv:2607.27581}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Grounding human motion in language, and language in motion, is a central step toward physical AI systems that can understand, generate, and communicate human behavior. Unified motion--language systems first coupled the two directions through a shared discrete motion codebook, but quantization limits generation quality. The strongest generators buy quality back at growing cost: stacked residual codebooks enlarge the representation; masked decoding stages, long autoregressive rollouts, and denoising chains of tens to hundreds of steps stretch inference; even the continuous-latent designs among them reach their latent only through an iterative diffusion head; and none of this decoding machinery serves understanding. We therefore propose MUGEN, a unified motion--language framework that pays neither cost: no codebook, one draw. A single adaptive-length autoencoder compresses any-length motion into a few continuous latent slots, the system's only motion representation: the language model generates them for text-to-motion and reads them back for motion understanding. Depth-routed hidden states let each slot read from the transformer depth it needs, and a calibrated head predicts a joint distribution over the full latent set, so a single draw carries the text-conditional, cross-slot variation a description permits. At a decoding cost of K language-model steps, one draw, and one decoder pass, MUGEN leads language-model baselines on FID on HumanML3D while raising retrieval precision above the real-motion reference under the standard evaluator, achieves the best CIDEr and BLEU@4 scores, and surpasses the discrete-token state of the art on every retrieval and alignment metric on SnapMoGen.

Figures

Figures reproduced from arXiv: 2607.27581 by Bingyang Wei, Bofan Li, Fangyi Li, Shangqian Gao, Xin Liu, Yukai Jin, Yusen Wu, Zhankai Ye.

Figure 1
Figure 1. Figure 1: The Adaptive-Length AutoEncoder (ALAE). The encoder compresses a motion of arbitrary length into exactly K continuous latent slots by letting learned latent queries cross-attend to a temporal memory. The decoder reconstructs the motion directly from those slots. serving as generation target, understanding input, and decoder input. 3 Methodology 3.1 Adaptive-Length AutoEncoder Without a Codebook Removing th… view at source ↗
Figure 2
Figure 2. Figure 2: The shared motion–language interface of MUGEN; flames mark trainable modules and snowflakes frozen [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Layer-selection preference of the two slots on [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Per-verb layer preference on HumanML3D: k0 (left) stays fixed on layer 10 for nearly every verb, whereas k1 (right) shifts its preferred layer with the semantics of the action—shallow for ballistic/contact verbs (crawl, punch, kick), mid-depth for locomotion (run, walk, jump), and deep for object/hand actions (pick, clap, wave). slot mean and diagonal scale, and the slot-pooled normalized states produce th… view at source ↗

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

53 extracted references · 2 canonical work pages

  1. [1]

    ECCV , pages =

    Guo, Chuan and Zuo, Xinxin and Wang, Sen and Cheng, Li , title =. ECCV , pages =. 2022 , isbn =. doi:10.1007/978-3-031-19833-5_34 , abstract =

    work page doi:10.1007/978-3-031-19833-5_34 2022

  2. [2]

    CVPR , pages=

    Generating diverse and natural 3d human motions from text , author=. CVPR , pages=

  3. [3]

    OpenAI blog , volume=

    Language models are unsupervised multitask learners , author=. OpenAI blog , volume=

  4. [4]

    Advances in neural information processing systems , volume=

    Neural discrete representation learning , author=. Advances in neural information processing systems , volume=

  5. [5]

    Proceedings of the 14th ACM international conference on Information and knowledge management , pages=

    A geometric interpretation and analysis of r-precision , author=. Proceedings of the 14th ACM international conference on Information and knowledge management , pages=

  6. [6]

    CVPR , pages=

    Generating human motion from textual descriptions with discrete representations , author=. CVPR , pages=

  7. [7]

    CVPR , pages=

    Momask: Generative masked modeling of 3d human motions , author=. CVPR , pages=

  8. [8]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    Stablemofusion: Towards robust and efficient diffusion-based motion generation framework , author=. Proceedings of the 32nd ACM International Conference on Multimedia , pages=

  9. [9]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Rethinking diffusion for text-driven human motion generation: Redundant representations, evaluation, and masked autoregression , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  10. [10]

    arXiv preprint arXiv:2507.09122 , year=

    Snapmogen: Human motion generation from expressive texts , author=. arXiv preprint arXiv:2507.09122 , year=

    arXiv

  11. [11]

    arXiv preprint arXiv:2209.14916 , year=

    Human motion diffusion model , author=. arXiv preprint arXiv:2209.14916 , year=

    Pith/arXiv arXiv

  12. [12]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Motiondiffuse: Text-driven human motion generation with diffusion model , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2024 , publisher=

    2024

  13. [13]

    Advances in Neural Information Processing Systems , volume=

    MotionGPT: Human motion as a foreign language , author=. Advances in Neural Information Processing Systems , volume=

  14. [14]

    arXiv preprint arXiv:2506.24086 , year=

    MotionGPT3: Human motion as a second modality , author=. arXiv preprint arXiv:2506.24086 , year=

    arXiv

  15. [15]

    2026 , isbn =

    Wang, Guocun and Liu, Kenkun and Lin, Jing and Song, Guorui and Li, Jian and Han, Xiaoguang , title =. 2026 , isbn =. doi:10.1609/aaai.v40i12.37936 , booktitle =

    work page doi:10.1609/aaai.v40i12.37936 2026

  16. [16]

    CVPR (CVPR) , year =

    VibeToken: Scaling 1D Image Tokenizers and Autoregressive Models for Dynamic Resolution Generations , author =. CVPR (CVPR) , year =

  17. [17]

    arXiv preprint arXiv:2601.07632 , year=

    GeoMotionGPT: Geometry-Aligned Motion Understanding with Large Language Models , author=. arXiv preprint arXiv:2601.07632 , year=

    arXiv

  18. [18]

    CVPR , pages=

    Executing your commands via motion diffusion in latent space , author=. CVPR , pages=

  19. [19]

    ECCV , pages=

    MotionLCM: Real-time controllable motion generation via latent consistency model , author=. ECCV , pages=. 2024 , organization=

    2024

  20. [20]

    Advances in Neural Information Processing Systems , volume=

    Motion-x: A large-scale 3d expressive whole-body human motion dataset , author=. Advances in Neural Information Processing Systems , volume=

  21. [21]

    ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model , year=

    Lu, Shunlin and Wang, Jingbo and Lu, Zeyu and Chen, Ling-Hao and Dai, Wenxun and Dong, Junting and Dou, Zhiyang and Dai, Bo and Zhang, Ruimao , booktitle=. ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model , year=

  22. [22]

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

    Go to zero: Towards zero-shot motion generation with million-scale data , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  23. [23]

    International Conference on Learning Representations , volume=

    LaMP: Language-motion pretraining for motion generation, retrieval, and captioning , author=. International Conference on Learning Representations , volume=

  24. [24]

    arXiv preprint arXiv:2411.19786 , year=

    MoTe: Learning Motion-Text Diffusion Model for Multiple Generation Tasks , author=. arXiv preprint arXiv:2411.19786 , year=

    Pith/arXiv arXiv

  25. [25]

    International conference on machine learning , pages=

    Perceiver: General perception with iterative attention , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  26. [26]

    International conference on learning representations , year=

    Categorical reparameterization with gumbel-softmax , author=. International conference on learning representations , year=

  27. [27]

    Advances in neural information processing systems , volume=

    Gans trained by a two time-scale update rule converge to a local nash equilibrium , author=. Advances in neural information processing systems , volume=

  28. [28]

    Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

    BLEU: a method for automatic evaluation of machine translation , author=. Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

  29. [29]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Cider: Consensus-based image description evaluation , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  30. [30]

    BERTScore: Evaluating Text Generation with BERT , author=

  31. [31]

    European conference on computer vision , pages=

    TEMOS: Generating diverse human motions from textual descriptions , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  32. [32]

    European Conference on Computer Vision , pages=

    MotionCLIP: Exposing human motion generation to clip space , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  33. [33]

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

    ReMoDiffuse: Retrieval-augmented motion diffusion model , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  34. [34]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    PhysDiff: Physics-guided human motion diffusion model , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  35. [35]

    CVPR , pages=

    MMM: Generative masked motion model , author=. CVPR , pages=

  36. [36]

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

    TMR: Text-to-motion retrieval using contrastive 3d human motion synthesis , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  37. [37]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    MG-MotionLLM: A unified framework for motion comprehension and generation across multiple granularities , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  38. [38]

    European Conference on Computer Vision , pages=

    BAMM: bidirectional autoregressive motion model , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  39. [39]

    International conference on machine learning , pages=

    Set transformer: A framework for attention-based permutation-invariant neural networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  40. [40]

    Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =

    Locatello, Francesco and Weissenborn, Dirk and Unterthiner, Thomas and Mahendran, Aravindh and Heigold, Georg and Uszkoreit, Jakob and Dosovitskiy, Alexey and Kipf, Thomas , title =. Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =. 2020 , isbn =

    2020

  41. [41]

    and Neumann, Mark and Iyyer, Mohit and Gardner, Matt and Clark, Christopher and Lee, Kenton and Zettlemoyer, Luke

    Peters, Matthew E. and Neumann, Mark and Iyyer, Mohit and Gardner, Matt and Clark, Christopher and Lee, Kenton and Zettlemoyer, Luke. Deep Contextualized Word Representations. Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018. doi:10...

    doi:10.18653/v1/n18-1202 2018

  42. [42]

    A Primer in BERT ology: What We Know About How BERT Works

    Rogers, Anna and Kovaleva, Olga and Rumshisky, Anna. A Primer in BERT ology: What We Know About How BERT Works. Transactions of the Association for Computational Linguistics. 2020. doi:10.1162/tacl_a_00349

    doi:10.1162/tacl_a_00349 2020

  43. [43]

    Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =

    Ho, Jonathan and Jain, Ajay and Abbeel, Pieter , title =. Proceedings of the 34th International Conference on Neural Information Processing Systems , articleno =. 2020 , isbn =

    2020

  44. [44]

    2022 , volume =

    Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bjorn , booktitle =. 2022 , volume =. doi:10.1109/CVPR52688.2022.01042 , url =

    arXiv 2022

  45. [45]

    , booktitle=

    Chang, Huiwen and Zhang, Han and Jiang, Lu and Liu, Ce and Freeman, William T. , booktitle=. MaskGIT: Masked Generative Image Transformer , year=

  46. [46]

    Autoregressive Image Generation using Residual Quantization , year=

    Lee, Doyup and Kim, Chiheon and Kim, Saehoon and Cho, Minsu and Han, Wook-Shin , booktitle=. Autoregressive Image Generation using Residual Quantization , year=

  47. [47]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Song, Yang and Dhariwal, Prafulla and Chen, Mark and Sutskever, Ilya , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  48. [48]

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

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

    2021

  49. [49]

    Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 2 , pages =

    Sohn, Kihyuk and Yan, Xinchen and Lee, Honglak , title =. Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 2 , pages =. 2015 , publisher =

    2015

  50. [50]

    2023 , volume =

    Zhong, Chongyang and Hu, Lei and Zhang, Zihao and Xia, Shihong , booktitle =. 2023 , volume =. doi:10.1109/ICCV51070.2023.00053 , url =

    arXiv 2023

  51. [51]

    2025 , url=

    Yang, An and Li, Anfeng and Yang, Baosong and Zhang, Beichen and Hui, Binyuan and others , journal=. 2025 , url=

    2025

  52. [52]

    2025 , url=

    Ben Allal, Loubna and Lozhkov, Anton and Bakouch, Elie and Bl\'azquez, Gabriel Mart\'in and Penedo, Guilherme and others , journal=. 2025 , url=

    2025

  53. [53]

    and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle=

    Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle=. 2022 , url=

    2022

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.