Pith. sign in

REVIEW 4 major objections 5 minor 23 references

TRiMM: Transformer-Based Rich Motion Matching for Real-Time multi-modal Interaction in Digital Humans

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read TRiMM combines a sliding-window transformer with a 9,143-action motion-matching graph to synthesize co-speech gestures at 120 fps with per-sentence latency around 0.15 s on consumer GPUs, while user ratings on ZEGGS and BEAT beat…

desk verdict Real-time gesture system with a plausible pipeline but a table that contradicts its own outperformance claim and an undocumented retrieval library; send to review but demand clarification. read the letter →

arxiv 2506.01077 v1 pith:XPVQRA5G submitted 2025-06-01 cs.GR cs.HC

Yueqian Guo , Tianzhao Li , Xin Lyu , Jiehaolin Chen , Zhaohan Wang , Sirui Xiao , Yurun Chen , Yezi He
show 2 more authors
Helin Li Fan Zhang
This is my paper
classification cs.GRcs.HC
keywords real-timegesturegenerationco-speechgesturesmotionmatchingtransformermulti-modalfusiondigitalhumansK-NNgraphsliding-windowautoregressive
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that co-speech gesture generation can be made fast enough for real-time digital-human interaction without sacrificing quality, and that its hybrid system achieves both. TRiMM replaces the slow iterative denoising of diffusion models with a sliding-window autoregressive transformer that predicts the next motion feature from text and audio, then looks up the best matching action in a prebuilt 9,143-node K-NN graph with duration filtering. On consumer GPUs it reports 120 fps inference, 0.14–0.19 seconds per sentence, and user ratings on the ZEGGS and BEAT datasets that beat diffusion, quantization, and retrieval baselines. A sympathetic reader would care because LLM-driven avatars that hold real conversations need gestures that appear as fast as speech, and current generation methods take seconds to minutes.

What carries the argument

The central object is a K-NN motion graph of 9,143 atomic actions, each carrying a 750-dimensional feature vector and a duration value. Retrieval starts at the nearest node to the previous action, explores the graph with a breadth-first search under a duration filter, and returns the node whose feature is closest to the transformer's predicted next-frame feature; quaternion spherical linear interpolation and cubic position interpolation blend the retrieved clips into a continuous motion. This graph is what lets the transformer output only a feature vector and still synthesize full, natural motion in real time.

What would settle it

Take the ground-truth BVH clips for the Wayne-subset BEAT test utterances and the ZEGGS clips, compute their 750-dimensional action features, and run the retrieval engine's search for each test sentence; if the best match is the ground-truth clip itself or a neighbor within a few percent of zero Euclidean distance, then the reported quality numbers come from test data returning through the library rather than from synthesis. The paper's released code and dataset split definitions make this check directly runnable.

Watch

Extended reading notes

Core claim

TRiMM claims to be the first system in its comparison to combine per-sentence latency under 0.2 seconds with user-rated gesture quality above existing diffusion-based, quantization-based, and pure-retrieval methods on both the ZEGGS and BEAT datasets. The central discovery is that a retrieval engine over a large atomic-action graph, fed by a transformer's predicted motion features, delivers the naturalness of learned generation and the speed of classical motion matching at once. The paper further claims ablations show that each module—the gated multimodal fusion, the divided space-time attention, and the K-NN motion graph—is necessary for the quality gain, while inference speed stays essentially unchanged.

Load-bearing premise

The load-bearing premise is that the 9,143-action motion library used by the retrieval engine contains no test or near-test clips; the paper explains how the graph is constructed but never states the provenance of the actions or the exclusion rule, so retrieval scores could be inflated if ground-truth motion is inside the library.

Editorial extensions

If this is right

  • A digital human can reply to a spoken sentence with synthesized gestures in roughly 0.15 seconds, making live conversational avatars practical in gaming, streaming, and virtual meetings.
  • Because the pipeline runs at 120 fps on a consumer laptop GPU, real-time gesture synthesis no longer requires offline precomputation or a dedicated server.
  • The ablations imply that neither pure generative nor pure retrieval alone is sufficient: removing the graph, the fusion gate, or the space-time attention each measurably degrades style, naturalness, or diversity.
  • Cross-dataset testing indicates the hybrid architecture transfers across speakers and conditions better than diffusion baselines (84% performance retention versus 52–67%).
  • The modular design lets the action library be swapped or expanded for new domains or characters without retraining the transformer.

Reading between the lines

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

  • The paper never states where the 9,143 atomic actions come from or whether any of them overlap the test clips it evaluates on; if the library contains near-duplicates of evaluation motions, retrieval could return essentially ground-truth gestures, inflating the reported diversity and user ratings.
  • Because synthesized gestures are concatenations of pre-recorded atoms, TRiMM's diversity is capped by the library content; a fairer comparison would measure how much novel motion it produces for actions not represented in the library.
  • A testable extension: exclude an entire style or speaker from the graph construction and measure FGD and user ratings, which would separate the transformer's predictive contribution from the library's coverage.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. TRiMM proposes a real-time co-speech gesture generation system that combines a Transformer-based autoregressive predictor with a large K-NN motion-matching library, plus a quaternion-based blending module, deployed in Unreal Engine. The paper claims 120 fps inference and 0.14–0.19 s per-sentence latency on consumer GPUs and reports subjective user-study scores and objective metrics on BEAT and ZEGGS that, the text argues, outperform diffusion, quantization, and retrieval baselines. The claimed contribution is a system that is both real-time and perceptually competitive with state-of-the-art generative methods.

Significance. If the claims were fully supported, this would be a practically valuable result: the paper demonstrates a working pipeline, provides a code repository, and includes a 24-participant user study with pairwise significance analysis. The central engineering achievement — sub-second per-sentence latency on an RTX 3060 while producing user-rated natural gestures — is credible and worth publishing. However, the current manuscript overstates its objective results: the paper's own Table 1 contradicts the claim that TRiMM outperforms the state of the art on FGD, and the provenance of the 9,143-action retrieval library is never tied to the train/test split, which is a load-bearing gap for a retrieval-augmented system.

major comments (4)
  1. [Table 1, ZEGGS block; §4.6.2; Abstract] The claim that TRiMM 'outperforms current state-of-the-art methods' is contradicted by the FGD column of Table 1. On ZEGGS, TRiMM reports FGD = 59011.57, while Diffsheg reports 10675.88 and ZeroEGGS reports 10202.83; since lower FGD is better, TRiMM is roughly five times worse than both. Yet the table bolds TRiMM's value as if it were the best, violating the table's own 'best-performing values are bold' rule. The text in §4.6.2 and the Abstract should be revised to qualify the outperformance claim (e.g., to latency and subjective metrics) and the table's bolding must be corrected.
  2. [§3.4.1, Eq. (7)–(9); §4.1; Appendix Table 2] The K-NN graph is constructed from 'a set of n actions' without stating where those actions come from. Section 4.1 describes segmenting BVH, text, and audio into clips, and the Appendix gives a BEAT train/test split (29 speakers train, Wayne test), but nothing connects graph construction to that split or excludes ZEGGS clips. If any evaluation or near-evaluation clip is in the 9,143-action library, the retrieval engine can return ground-truth motion, which would inflate subjective ratings and also affect the diversity and beat-alignment numbers. The paper must explicitly state that the library is built only from the training partition, exclude the BEAT test speaker and the ZEGGS evaluation clips, and either report a formal overlap check or provide a leakage analysis.
  3. [§4.6.3, Table 1 BEAT block] The ablation claims do not match the numbers in Table 1. The text says 'Motion Graph Removal (MGA) caused the sharpest style degradation (1.11→-0.64 on BEAT)', but the same table shows TRiMM MFA with style appropriateness -0.75, which is a larger drop from 1.11. The adjacent text also says MFA caused 'marginally increasing FGD', while the ZEGGS row shows FGD decreasing from 59011.57 to 58901.43. The ablation conclusions and the footnote need to be revised to be consistent with the reported numbers.
  4. [§4.4.1 and Appendix 'User study details'] The paper gives two incompatible descriptions of the subjective-evaluation sampling protocol. Section 4.4.1 states that 'randomly extracts 20 data of a role in the BEAT dataset and 20 data of a style in the zeggs dataset for each participant', whereas the Appendix states that 'each subgroup received 5 randomly sampled clips from the BEAT dataset and 5 randomly selected styles from the ZEGGS dataset'. Since the subjective results carry much of the paper's outperformance claim, the sampling protocol must be stated unambiguously and consistently, and the reported number of stimuli per participant must be reconciled.
minor comments (5)
  1. [Abstract, §1, §2, §4.6.2, Table 1] The reported latency numbers are inconsistent: the Abstract says 0.15 s, §1 says 0.159 s and 0.13 s, Table 1 gives 0.14 s (ZEGGS) and 0.19 s (BEAT), and §4.6.2 quotes Diffsheg's AITS as 0.22 while Table 1 shows 0.32. Please harmonize these values.
  2. [§1 and §5] The phrase 'raises motion diversity by 2 times' and 'twice that of traditional motion capture systems' is not tied to any specific baseline or measurement. Please define the comparison explicitly or remove the quantitative claim.
  3. [§3.1] The feature-extractor names are inconsistently capitalized ('WAVE2vec2' vs. 'wav2vec') and the audio feature dimension is given as 2048 in §3.1 but as 512 in §3.2.1. Please correct these details.
  4. [Throughout] The writing contains many grammatical and typographical issues, e.g., 'cube interpolation' for cubic interpolation, 'human like nature', and 'multi-modal' capitalization inconsistencies. A thorough language edit is needed.
  5. [§4.6.4] The cross-dataset generalization claim ('maintains 84% of original performance') is not backed by any table or statistical test, only by a single pair of numbers in text. Please provide the supporting data or qualify the claim.

Circularity Check

0 steps flagged · score 2.0 of 10

No equation-level circularity; the only self-citation is a non-load-bearing methodological borrowing for the user-study protocol, and the main claims are empirical measurements against external baselines.

full rationale

This is an empirical systems paper rather than a derivation, so there is no chain of equations whose output is equivalent to its input. The transformer is trained with MSE loss in a 750-dimensional PCA feature space (Eq. 13), and the motion-matching engine retrieves nearest neighbors from a precomputed k-NN graph (Eqs. 7-9); the reported speed, FGD, diversity, beat-align, and user-study numbers are measured outcomes of that pipeline, not consequences forced by the loss or by the graph construction. The central latency claim (AITS 0.14-0.19 s) is a direct measurement, and the quality claims are comparisons against externally published baselines (Diffsheg, ZeroEGGS, DiffuseGesture, QPGesture), so they do not reduce by construction to the training objective. The only self-citation I found is in Section 4.4.1, where the subjective evaluation 'draws on the papers of [Zhang et al. 2024c]' and adopts the same paired-comparison protocol and seven-point scale. That prior TVCG paper shares authors with the present work, so it is a minor self-citation, but it is methodological rather than load-bearing: the preference scores are newly collected from 24 participants, and neither the speed result nor the graph-retrieval result depends on that citation. The appendix also states that user-study samples were excluded from training and validation, which addresses leakage for the subjective study. The remaining provenance concern - whether the 9,143-action library overlaps the BEAT test split or ZEGGS corpus - is a data-leakage and correctness risk, not a circularity, because no passage shows that the retrieval output is defined in terms of the evaluation target. I therefore assign 2 rather than 0 only to flag the non-load-bearing self-citation; there is no specific circular reduction to quote.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim depends on the standard components listed above; no new physical or mathematical entities are introduced.

free parameters (7)
  • window_size_N
    Chosen by hand; not reported as a number in the text; determines how much context the autoregressive model sees.
  • K_neighbors = 10
    Number of nearest neighbors used to build the K-NN graph, stated in Figure 4.
  • top_k_search
    Number of neighboring nodes considered at the start of retrieval in Algorithm 1; value not reported.
  • duration_threshold_tau
    Hand-set minimum action duration used as a filter in Algorithm 1; value not reported.
  • PCA_and_projection_dims = 2048 audio, 750 motion, 1024 fusion, 2048 final
    Feature dimensionalities chosen for standardization in Sections 3.1 and 3.4.1; they define the spaces used for retrieval and evaluation.
  • training_hyperparameters = batch 256, lr 1e-4, epochs 10000, decay 0.999
    Hand-chosen training settings reported in Section 4.2; the central results depend on them.
  • learned_network_weights = trained on BEAT with MSE
    Projection, gating, self-attention, and output-layer parameters are fitted to data; they implement the entire prediction mapping.
assumptions (5)
  • domain assumption Pretrained Wav2Vec2 and BERT features carry enough audio and semantic information for gesture prediction.
    Section 3.1 uses these extractors as the only input representation; if they miss prosodic or semantic cues, the whole pipeline is limited.
  • domain assumption Euclidean distance in the 750-dimensional PCA motion space is a valid similarity measure for matching and for FGD evaluation.
    Section 3.4 and Section 4.4.2 define retrieval and FGD on this metric; no validation that this space preserves perceptual motion similarity.
  • domain assumption Minimizing MSE between predicted and true 750-d action features is sufficient for natural gesture quality.
    Section 3.6 uses MSE only; no perceptual or adversarial loss is used.
  • domain assumption Evaluation clips are not present in the 9,143-action motion library.
    Section 3.4 constructs the K-NN graph from action features but never states the training or test provenance; retrieval can leak ground truth if violated.
  • domain assumption Anonymous random assignment and 24 participants yield reliable preference estimates across tasks.
    Section 4.4.1 reports 24 volunteers; no power analysis or inter-rater reliability is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TRiMM: Transformer-Based Rich Motion Matching for Real-Time multi-modal Interaction in Digital Humans." pith.science (2026). https://pith.science/paper/XPVQRA5G

@misc{pith2026250601077,
  author       = {Pith},
  title        = {Pith review of: TRiMM: Transformer-Based Rich Motion Matching for Real-Time multi-modal Interaction in Digital Humans},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XPVQRA5G}},
  note         = {Machine review of arXiv:2506.01077}
}
read the original abstract

Large Language Model (LLM)-driven digital humans have sparked a series of recent studies on co-speech gesture generation systems. However, existing approaches struggle with real-time synthesis and long-text comprehension. This paper introduces Transformer-Based Rich Motion Matching (TRiMM), a novel multi-modal framework for real-time 3D gesture generation. Our method incorporates three modules: 1) a cross-modal attention mechanism to achieve precise temporal alignment between speech and gestures; 2) a long-context autoregressive model with a sliding window mechanism for effective sequence modeling; 3) a large-scale gesture matching system that constructs an atomic action library and enables real-time retrieval. Additionally, we develop a lightweight pipeline implemented in the Unreal Engine for experimentation. Our approach achieves real-time inference at 120 fps and maintains a per-sentence latency of 0.15 seconds on consumer-grade GPUs (Geforce RTX3060). Extensive subjective and objective evaluations on the ZEGGS, and BEAT datasets demonstrate that our model outperforms current state-of-the-art methods. TRiMM enhances the speed of co-speech gesture generation while ensuring gesture quality, enabling LLM-driven digital humans to respond to speech in real time and synthesize corresponding gestures. Our code is available at https://github.com/teroon/TRiMM-Transformer-Based-Rich-Motion-Matching

Figures

Figures reproduced from arXiv: 2506.01077 by the authors.

Figure 1
Figure 1. The system trains a Transformer model using text, speech, and motion feature data, performs real-time multi-modal [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. This system processes text and audio inputs through dedicated embedding layers, combines them via a fusion [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. This figure illustrates the autoregressive prediction process. And the connection between autoregressive prediction, [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: K-NN Graph Visualization after t-SNE Dimensionality Reduction. The graph consists of 9,143 action nodes, where [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: The real-time motion is inferred from JSON signals transmitted through the LiveLink plugin, then rendered in Unreal [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Visual comparisons of motion generation results, GT(ground truth) represents the original motion data, and the other [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: The Average Inference Time per Sentence of each module in TRiMM. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: The mean rating of each metric for each approach across the two datasets in comparative experiments. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: The mean rating of each metric for each approach across the two datasets in ablation experiments. [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: userstudy program UI In our user study, we adopted a 7-point Likert scale to measure participants’ subjective preferences. According to Nunnally’s seminal work, scale reliability tends to plateau around seven response categories, with marginal gains observed beyond el…
Figure 11
Figure 11. Figure 11: Pairwise significance heatmaps of gesture generation methods on BEAT dataset [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]
Figure 12
Figure 12. Figure 12: Pairwise significance heatmaps of gesture generation methods on ZEGGS dataset [PITH_FULL_IMAGE:figures/full_fig_p024_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 10 canonical work pages

  1. [1]

    Sicheng Yang, Zhiyong Wu, Minglei Li, Zhensong Zhang, Lei Hao, Weihong Bao, Ming Cheng, and Long Xiao. 2023b. DiffuseStyleGesture: Stylized Audio-Driven Co-Speech Gesture Generation with Diffusion Models. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence (IJCAI-23) . IJCAI. https://www.ijcai.org/proceedings/2023...

    arXiv 2023

  2. [6]

    In Special Interest Group on Computer Graphics and Interactive Techniques Conference Proceedings

    A Motion Matching-based Framework for Controllable Gesture Synthesis from Speech. In Special Interest Group on Computer Graphics and Interactive Techniques Conference Proceedings . ACM, Vancouver BC Canada, 1–9. doi:10.1145/3528233.3530750 Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter

    arXiv

  3. [12]

    In 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Action-Conditioned 3D Human Motion Synthesis with Transformer VAE. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE, Montreal, QC, Canada, 10965–10975. doi:10.1109/ICCV48922.2021.01080 Mathis Petrovich, Michael J Black, and Gül Varol

    arXiv 2021

  4. [13]

    doi:10.21437/Interspeech.2019-1873 Sinan Sonlu, Bennie Bendiksen, Funda Durupinar, and Uğur Güdükbay

    3465–3469. doi:10.21437/Interspeech.2019-1873 Sinan Sonlu, Bennie Bendiksen, Funda Durupinar, and Uğur Güdükbay

    doi:10.21437/interspeech.2019-1873 2019

  5. [14]

    http://arxiv.org/abs/2407.10993 arXiv:2407.10993 [cs]

    The Effects of Embodiment and Personality Expression on Learning in LLM-based Educational Agents. http://arxiv.org/abs/2407.10993 arXiv:2407.10993 [cs]. , Vol. 1, No. 1, Article . Publication date: June . • Yueqian Guo, Tianzhao Li, Xin Lyu, Jiehaolin Chen, Zhaohan Wang, Sirui Xiao, Yurun Chen, Yezi He, Helin Li, and Fan Zhang Xiangjun Tang, He Wang, Bo H...

    arXiv

  6. [15]

    ACM Transactions on Graphics (TOG) 41, 4 (2022), 1–10

    Real-time controllable motion transition for characters. ACM Transactions on Graphics (TOG) 41, 4 (2022), 1–10. Guy Tevet, Sigal Raab, Brian Gordon, Yonatan Shafir, Daniel Cohen-Or, and Amit H. Bermano

    work page 2022

  7. [16]

    doi:10.48550/arXiv.2209.14916 arXiv:2209.14916 [cs]

    Human Motion Diffusion Model. doi:10.48550/arXiv.2209.14916 arXiv:2209.14916 [cs]. Hongyu Wan, Jinda Zhang, Abdulaziz Arif Suria, Bingsheng Yao, Dakuo Wang, Yvonne Coady, and Mirjana Prpa

    arXiv doi:10.48550/arxiv.2209.14916

  8. [17]

    In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI EA ’24)

    Building LLM-based AI Agents in Social Virtual Reality. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI EA ’24). Association for Computing Machinery, New York, NY, USA, Article 65, 7 pages. doi:10.1145/3613905.3651026 Sen Wang, Jiangning Zhang, Weijian Cao, Xiaobin Hu, Moran Li, Xiaozhong Ji, Xin ...

    arXiv

Show all 23 references
  1. [18]

    CoRR abs/2403.02905 (2024)

    MMoFusion: Multi-modal Co-Speech Motion Generation with Diffusion Model. CoRR abs/2403.02905 (2024). doi:10.48550/ARXIV.2403.02905 arXiv:2403.02905 Qi Wu, Yubo Zhao, Yifan Wang, Xinhang Liu, Yu-Wing Tai, and Chi-Keung Tang

    2024 doi arXiv

  2. [21]

    ACM Transactions on Graphics 41, 6 (Dec

    ControlVAE: Model-Based Learning of Generative Controllers for Physics- Based Characters. ACM Transactions on Graphics 41, 6 (Dec. 2022), 1–16. doi:10.1145/3550454.3555434 arXiv:2210.06063 [cs]. Fan Zhang, Zhaohan Wang, Xin Lyu, Siyuan Zhao, Mengjian Li, Weidong Geng, Naye Ji,...

    2024

  3. [22]

    In 2023 IEEE/CVF International Conference on Computer Vision (ICCV)

    ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV) . IEEE, Paris, France, 364–373. doi:10.1109/ICCV51070.2023.00040 Mingyuan Zhang, Daisheng Jin, Chenyang Gu, Fangzhou Hong, Zhongang Cai, Jingfang Huang,...

    2023

  4. [23]

    IEEE Transactions on Multimedia 24 (2021), 1909–1921

    Approximate k-NN graph construction: a generic online approach. IEEE Transactions on Multimedia 24 (2021), 1909–1921. Wenyang Zhou, Zhiyang Dou, Zeyu Cao, Zhouyingcheng Liao, Jingbo Wang, Wenjia Wang, Yuan Liu, Taku Komura, Wenping Wang, and Lingjie Liu

    2021

  5. [24]

    In Computer Vision – ECCV 2024, Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and Gül Varol (Eds.)

    EMDM: Efficient Motion Diffusion Model for Fast and High-Quality Motion Generation. In Computer Vision – ECCV 2024, Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and Gül Varol (Eds.). Vol. 15060. Springer Nature Switzerland, Cham, 18–38. doi:10.1...

    2024 doi

  6. [30]

    https://proceedings.neurips.cc/paper/2017/hash/8a1d694707eb0fefe65871369074926d-Abstract.html Daniel Holden, Oussama Kanoun, Maksym Perepichka, and Tiberiu Popa

    Curran Associates, Inc. https://proceedings.neurips.cc/paper/2017/hash/8a1d694707eb0fefe65871369074926d-Abstract.html Daniel Holden, Oussama Kanoun, Maksym Perepichka, and Tiberiu Popa

    2017

  7. [32]

    Curran Associates, Inc. https://proceedings.neurips.cc/paper/2019/hash/ 7ca57a9f85a19a6e4b9a248c1daca185-Abstract.html Haiyang Liu, Zihao Zhu, Naoya Iwamoto, Yichen Peng, Zhengqing Li, You Zhou, Elif Bozkurt, and Bo Zheng

    2019

  8. [2010]

    In 2010 22nd International Teletraffic Congress (LTC 22)

    On the bias of BFS (breadth first search). In 2010 22nd International Teletraffic Congress (LTC 22). IEEE, 1–8. Chang-Hung Lee, Cheng-Ru Lin, and Ming-Syan Chen

    2010

  9. [2016]

    GDC 2016 Presentation (2016)

    Motion Matching and The Road to Next-Gen Animation. GDC 2016 Presentation (2016). https://archive.org/details/ GDC2016Clavet Wenxun Dai, Ling-Hao Chen, Jingbo Wang, Jinpeng Liu, Bo Dai, and Yansong Tang

    2016

  10. [2019]

    In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers)

    Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers) . 4171–4186. Sa...

    2019

  11. [2020]

    ACM Trans

    Learned motion matching. ACM Trans. Graph. 39, 4 (Aug. 2020), 53:53:1–53:53:12. doi:10.1145/3386569.3392440 Korrawe Karunratanakul, Konpat Preechakul, Supasorn Suwajanakorn, and Siyu Tang

    2020

  12. [2021]

    ACM Transactions on Graphics 40, 4 (Aug

    ChoreoMaster: choreography- oriented music-driven dance synthesis. ACM Transactions on Graphics 40, 4 (Aug. 2021), 1–13. doi:10.1145/3450626.3459932 Xin Chen, Biao Jiang, Wen Liu, Zilong Huang, Bin Fu, Tao Chen, and Gang Yu

    2021

  13. [2022]

    Nature Reviews Methods Primers 2, 1 (2022),

    Principal component analysis. Nature Reviews Methods Primers 2, 1 (2022),

    2022

  14. [2024]

    In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    DiffSHEG: A Diffusion-Based Approach for Real-Time Speech-Driven Holistic 3D Expression and Gesture Generation. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, Seattle, WA, USA, 7352–7361. doi:10.1109/CVPR52733.2024.00702 Kang Chen, Zhipeng...

    2024

  15. [2025]

    doi:10.48550/arXiv.2503

    MotionStreamer: Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space. doi:10.48550/arXiv.2503. 15451 arXiv:2503.15451 [cs] version:

    doi arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.