Pith. sign in

REVIEW 5 major objections 5 minor 98 references

MRBench: A Comprehensive Benchmark for Human Motion-Text Retrieval

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper introduces MRBench, a heterogeneous, multi-granularity benchmark for human motion-text retrieval, and a granularity-aware model that improves mixed-granularity retrieval without sacrificing standard-caption performance.

desk verdict A genuinely useful benchmark proposal for motion-text retrieval, but the lack of curation transparency and missing data release keep it from being trustworthy yet. read the letter →

arxiv 2608.07993 v1 pith:WBRFUAQA submitted 2026-08-08 cs.CV

Fulong Liu , Liang Xu , Chengqun Yang , Yuhao Zhang , Yichao Yan , Xiaokang Yang This is my paper · ORCID
classification cs.CV
keywords humanmotion-textretrievalbenchmarkmulti-granularcaptionscross-modalalignmentgranularity-awaremotion-languagedatasetscontrastivelearningdatasetcuration
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 argues that existing motion-text retrieval benchmarks, such as HumanML3D and KIT-ML, largely measure how well models match repetitive indoor motion-capture captions rather than true cross-modal understanding. It reports that in those datasets many test queries share identical text with other gallery motions, so even correct retrievals are often counted as wrong. To address this, the authors construct MRBench, a benchmark of 3,390 motions drawn from motion capture, in-the-wild video, synthetic video, and generative models, covering 118 categories, with each motion labeled by concise, standard, and fine-grained captions for a total of 10,170 captions. They show that current retrieval models drop sharply on MRBench and are sensitive to caption granularity. They also propose a lightweight granularity-aware model that freezes a standard-caption-aligned backbone and adds granularity-specific extractors and adapters, which improves fine-grained and mixed-granularity retrieval without lowering standard-caption performance.

What carries the argument

The load-bearing object is the benchmark itself, MRBench, containing 3,390 motions standardized to a unified skeleton at 20 FPS with 10,170 captions. It is built by a four-stage pipeline: text-guided filtering using rule-based checks and large-language-model scoring; taxonomy-guided balanced sampling into 7 coarse, 33 middle-level, and 118 fine-grained categories with embedding-based deduplication; semantic alignment verification using a vision-language model that checks whether the primary motion described in a caption is visible in sampled frames; and multi-granular rewriting, where the same vision-language model generates concise and fine-grained variants that are then manually validated. The proposed model uses a frozen dual encoder trained on standard captions as an alignment anchor, then trains lightweight attention-pooling motion extractors and text-projection adapters on LLM-rewritten concise and fine-grained captions as pseudo-supervision. At inference, the known query granularity selects a branch, and for non-standard granularities the global and adapted cosine similarities are fused with a granularity-specific weight so that scores remain comparable across all description levels.

What would settle it

Take a random sample of MRBench motions and have independent human annotators, who have not seen the construction pipeline, judge whether each concise, standard, and fine-grained caption matches the motion; if a substantial fraction of captions are judged mismatched or ambiguous, the claim that MRBench provides reliable, motion-verified, unambiguous captions is false. Alternatively, recompute the category distribution after merging semantically identical labels, and if it becomes as concentrated as HumanML3D's distribution, the balance claim fails.

Watch

Extended reading notes

Core claim

The central claim is that a useful motion-text retrieval benchmark must vary both the motion distribution and the textual granularity to expose genuine alignment ability. The paper claims MRBench is the first such benchmark, combining heterogeneous motion sources with balanced category coverage and unique, motion-verified, multi-granular captions. On this benchmark, the authors observe a substantial cross-dataset generalization gap: methods trained on HumanML3D perform far worse on MRBench, and all evaluated baselines are sensitive to whether the query is concise, standard, or fine-grained. They further claim that their granularity-aware model, which keeps a frozen standard-caption-aligned dual encoder and learns granularity-specific attention-based motion extractors and text adapters with calibrated score fusion, improves fine-grained and mixed-granularity retrieval while preserving standard-caption retrieval exactly.

Load-bearing premise

The benchmark's reliability rests on the assumption that the automated filtering and verification pipeline, plus a small unquantified manual review, correctly judged every caption to match its motion and to be unambiguous.

Editorial extensions

If this is right

  • On MRBench, models trained on HumanML3D lose a large share of their recall, so in-domain scores on existing benchmarks overstate robustness to heterogeneous motion and text.
  • Retrieval quality depends strongly on query granularity: concise and fine-grained captions are harder than standard captions for all tested baselines.
  • Mixed-granularity retrieval requires score calibration across branches; removing that calibration degrades motion-to-text ranking substantially.
  • Training on MRBench-Train transfers better to MRBench than training on KIT-ML, HumanML3D, or MotionMillion, indicating that motion diversity and annotation discriminability matter more than raw data scale.
  • The granularity-aware model improves fine-grained and mixed-granularity retrieval without changing standard-caption results, demonstrating that robustness to non-standard queries can be added without sacrificing the original alignment.

Reading between the lines

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

  • A natural next step is to publish inter-annotator agreement statistics from the manual validation pass; until then, the paper's claim that all captions are motion-verified cannot be independently checked.
  • The benchmark's usefulness depends on caption uniqueness, and the paper reports that after merging mirrored pairs, 19.6% and 5.9% of existing benchmark queries remain ambiguous while MRBench's rate is roughly zero; auditing that comparison directly would test the benchmark's discriminative advantage.
  • The granularity-aware split-branch recipe could transfer to image-text or video-text retrieval, where caption specificity also changes retrieval difficulty.
  • Because the concise and fine-grained pseudo-labels come from a large language model, part of the model's gain may reflect matching LLM rewriting style rather than deeper motion understanding; rewriting the same motions with a different model and measuring retention would separate those effects.
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

5 major / 5 minor

Summary. The paper introduces MRBench, a motion-text retrieval benchmark containing 3,390 motion sequences from MoCap, in-the-wild video, synthetic video, and generative-model sources, with 118 fine-grained categories and three caption granularities (concise, standard, fine-grained), totaling 10,170 captions. The benchmark is built through a multi-stage pipeline: GPT-5.5-based caption filtering, Qwen-based taxonomy assignment, balanced sampling, Gemini-based semantic alignment verification, Gemini-based multi-granular rewriting, and manual validation. The paper also proposes a granularity-aware retrieval model that freezes a standard-caption-aligned dual encoder and trains lightweight motion extractors and text adapters on LLM-rewritten concise and fine-grained captions, with score fusion at inference. The experiments evaluate several baselines on MRBench under single-granularity and mixed-granularity protocols, reporting cross-dataset generalization gaps and granularity sensitivity, and the proposed model is shown to improve fine-grained retrieval while preserving standard-caption performance.

Significance. If the benchmark's curation claims are valid, MRBench addresses a real gap in motion-text retrieval evaluation, which currently relies on homogeneous, imbalanced datasets with repetitive captions. The statistical analysis of HumanML3D and KIT-ML in Fig. 2 is useful and gives concrete evidence of the ambiguity problems in existing benchmarks. The multi-source composition and the three-granularity annotation scheme are sensible design choices, and the proposed model, which anchors on a frozen standard-aligned encoder, is simple and well motivated. The cross-dataset generalization results are falsifiable and potentially valuable for the community. However, the benchmark's foundational reliability, the proposed model's gains, and the claimed score comparability in mixed-granularity retrieval cannot currently be assessed because the curation pipeline lacks quantified acceptance criteria, the evaluation may be confounded with LLM caption style, and no data or code release is stated.

major comments (5)
  1. [Benchmark Construction] The central claim that MRBench contains 'motion-verified, well-aligned' captions is not independently checkable because no quantitative acceptance criteria are reported for the curation stages. 'Text-Guided Candidate Filtering' lists GPT-5.5 scores (visual noise, object noise, motion specificity, low ambiguity) but gives neither thresholds nor score distributions; 'Semantic Alignment and Disambiguation' reports no SSAE pass rate, no alignment acceptance threshold, and no rationale-quality check for the Gemini-3-pro-preview verdicts; and the manual review in 'Multi-granular Text Expansion' provides no inter-annotator agreement, correction statistics, or per-granularity verification counts. Because Tables 1-3 evaluate retrieval on these captions, the validity of every headline result is currently unverified.
  2. [Granularity-Aware Retrieval Model (Training and Inference)] The fine-grained evaluation is exposed to a style-circularity risk. The MRBench fine-grained captions are produced by Gemini-3-pro-preview from standard captions plus alignment rationale, while the proposed model's concise and fine-grained branches are trained with pseudo-labels generated by Qwen2.5-7B-Instruct from the same standard captions. Reported fine-grained gains may reflect adaptation to LLM paraphrase statistics rather than genuine motion understanding. This is a testable concern: report results on held-out human-written fine-grained queries or on LLM-generated captions with style perturbed, and report how model performance changes with caption style statistics. Note also that Gemini is used for both SSAE verification and rewriting, so any systematic bias in its motion-visibility judgments is shared between the verification and the generated captions.
  3. [Granularity-Aware Inference, Eq. (7)] The statement that Eq. (7) 'makes fused scores across different textual granularities directly comparable' is stronger than the formula supports. Dividing by (1+alpha_g) applies a fixed per-granularity scale but does not correct for different location or shape of the global similarity distributions across concise, standard, and fine-grained queries. In the Mixed3 protocol this may leave per-granularity offsets in the ranking. Please report score distributions by granularity on the validation split, add an explicit per-granularity calibration if needed, and demonstrate that Mixed3 rankings are robust to monotone per-granularity transformations.
  4. [Benchmark Statistics and Granularity-Aware Retrieval Model] There is a contradiction in the training-data description. The 'Benchmark Statistics' section says that MRBench-Train 'follows the same annotation protocol' as MRBench, which includes multi-granular rewriting, but the method section says that concise and fine-grained variants are expanded by Qwen2.5-7B-Instruct because the training set consists of standard-caption pairs only. Please clarify whether MRBench-Train contains multi-granular captions or only standard captions, and specify exactly what 'same annotation protocol' means.
  5. [Benchmark Statistics] The paper provides no data or code availability statement, release URL, or licensing information for MRBench. A benchmark paper's central artifact must be publicly available for the retrieval scores in Tables 1-3 to be reproducible and for the claimed community testbed role to be realized; please add a clear availability statement.
minor comments (5)
  1. [Table 1] In Table 1, where all models are trained on H3D, the proposed model is absent from the H3D Standard block; since the model is reported to preserve standard-caption performance, please include this entry or state explicitly why it is omitted.
  2. [Table 2] In Table 2, the MotionMillion block does not include SGAR although SGAR is listed as a baseline; state whether the results are unavailable or were omitted.
  3. [Figure 3] The label 'Motion Diverity' in Figure 3 should be corrected to 'Motion Diversity'.
  4. [Eq. (2)] In Eq. (2), specify the softmax axis and state the dimensions of Wk and Wv so that the attention pooling over the motion token sequence is unambiguous.
  5. [Figure 2(c)] Figure 2(c) references 'mirrored pairs' without a definition; please explain how mirrored pairs are identified and merged for each dataset, since the merging rule affects the reported ambiguity percentages.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark construction and the granularity-aware model are derived from external data and explicit equations, with no fitted parameter or self-citation serving as the load-bearing reduction.

full rationale

The paper's central derivations are self-contained against external benchmarks. The MRBench statistics (category concentration, caption length, text collision rates) are computed directly from the official test splits of HumanML3D, KIT-ML, and MRBench, so the claimed generalization gap is an independently measurable quantity rather than an artifact of the paper's own assumptions. The granularity-aware model is defined by explicit equations: Eq. (1) gives the frozen global branch, Eqs. (2)-(4) define the granularity-specific extractors and adapters, Eq. (5) is an InfoNCE loss over LLM-rewritten pseudo-labels, and Eqs. (6)-(7) specify a convex score fusion whose weight alpha is selected on the validation split. None of these equations reduces to the benchmark's own labels: the training pseudo-captions come from Qwen2.5-7B-Instruct, while the test captions come from Gemini-3-pro-preview, and the model is evaluated on retrieval rankings, not on regenerating benchmark captions. The SSAE alignment procedure is adopted from the external HY-Motion work, not from a same-author uniqueness theorem. The use of LLMs for filtering, verification, and rewriting is a validity and reproducibility concern because acceptance thresholds and inter-annotator agreement are not reported, but that is an empirical limitation, not a circular derivation. No fitted parameter is renamed as a prediction, and no load-bearing premise is justified solely by self-citation. The paper's ablation study, even with its unusual tied entries, does not exhibit any equation-level equivalence between what is fit and what is claimed as a result. Accordingly, the appropriate circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the quality of the source corpus and on LLM-based curation judgments. Filtering and alignment thresholds are not disclosed, one fusion weight alpha_g is fit on validation, and the benchmark's category balance is defined by a hand-chosen per-category range. No genuinely new physical or representational entity is introduced beyond the dataset and adapter modules.

free parameters (5)
  • Granularity fusion weight alpha_g = Not reported (selected on validation split)
    Eq. (7) weights global and adapted similarities for concise and fine-grained branches; chosen on validation and fixed at test time.
  • GPT-5.5 filtering thresholds = Not reported
    Text-guided candidate filtering uses LLM scores for motion specificity, visual noise, object noise, and ambiguity, with unspecified cutoffs that determine which candidates enter the benchmark.
  • SSAE alignment acceptance threshold = Not reported
    Gemini-3-pro-based Semantic Alignment Evaluation decides which motion-text pairs are retained; the decision rule and confidence threshold are not specified.
  • bge-en-icl deduplication threshold = Not reported
    Semantic deduplication of captions within dense categories relies on embedding similarity with no stated cutoff, affecting balance and uniqueness.
  • Per-category sample range 10 to 60 = 10 to 60, median 42
    Taxonomy-guided balanced sampling targets 10 to 60 samples per category; this design choice defines what balanced means in the benchmark.
assumptions (4)
  • domain assumption ViMoGen-228K is a reliable source of diverse, correctly reconstructed 3D motions and semantically aligned captions.
    The entire benchmark is sampled from ViMoGen-228K; errors in source motions or captions propagate to MRBench.
  • domain assumption LLM and VLM judgments from GPT-5.5, Qwen2.5-7B-Instruct, and Gemini-3-pro-preview correctly identify motion semantics and motion-text alignment.
    These models are used for caption filtering, taxonomy assignment, alignment verification, and multi-granular rewriting; no human inter-annotator agreement is reported.
  • domain assumption The HY-Motion hierarchical taxonomy is a valid and complete categorization of human motions.
    Taxonomy-guided sampling and category-balance claims rely on this taxonomy; incomplete categories would bias the 118 fine-grained categories claim.
  • standard math Contrastive dual-encoder training with InfoNCE produces a valid cross-modal retrieval space.
    The base retrieval model follows standard practice from Oord et al. 2018 and Radford et al. 2021, assumed as background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MRBench: A Comprehensive Benchmark for Human Motion-Text Retrieval." pith.science (2026). https://pith.science/paper/WBRFUAQA

@misc{pith2026260807993,
  author       = {Pith},
  title        = {Pith review of: MRBench: A Comprehensive Benchmark for Human Motion-Text Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WBRFUAQA}},
  note         = {Machine review of arXiv:2608.07993}
}
read the original abstract

Human motion-text retrieval provides a rigorous means of assessing cross-modal alignment. Prevailing benchmarks are dominated by homogeneous indoor motions, imbalanced motion distributions, and oversimplified, repetitive texts, which hinder the reliable measurement of cross-domain and cross-granularity alignment. We thus introduce MRBench, a comprehensive motion-text retrieval benchmark featuring heterogeneous motions, broad and balanced category coverage, and reliable, discriminative, multi-granular descriptions. MRBench is constructed through a meticulously designed multi-stage data curation pipeline, which filters and balances candidates, verifies unambiguous semantic alignment, and generates motion-grounded descriptions at multiple granularities. The resulting benchmark contains 3,390 motions drawn from motion capture, in-the-wild videos, synthetic videos, and motion generative models, covering 118 fine-grained categories. Each motion is paired with concise, standard, and fine-grained descriptions, yielding 10,170 captions. Extensive evaluations of representative retrieval baselines on MRBench reveal a substantial cross-dataset generalization gap and pronounced sensitivity to query granularity. We propose a lightweight granularity-aware model anchored at a frozen standard-caption-aligned retrieval model. LLM-based concise and fine-grained captions provide pseudo-supervision for extra-branch granularity-specific motion extractors and text adapters. For inference, granularity-aware score fusion integrates global and adapted similarities while strictly maintaining score comparability across all description levels. The resulting model improves mixed-granularity retrieval without compromising standard-caption performance. We believe that our MRBench provides a comprehensive testbed for advancing motion-language alignment evaluation.

Figures

Figures reproduced from arXiv: 2608.07993 by the authors.

Figure 1
Figure 1. Overview of MRBench. MRBench features diverse motion sources, broad motion-category coverage, and multi￾granularity text descriptions, enabling a more comprehensive evaluation of human motion-text retrieval. Based on the ViMoGen-228K (Lin et al. 2025) dataset, we first remove generic, repetitive, and non-kinematic captions through rule-based filtering and LLM-based semantic scor￾ing, and then perform taxonomy-guided… view at source ↗
Figure 2
Figure 2. Statistical comparisons between MRBench and existing benchmarks, computed on the official test splits. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of the MRBench construction pipeline. Starting from ViMoGen-228K, we perform retrieval-oriented caption [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Overview of our granularity-aware retrieval framework. Standard captions are rewritten into concise and fine￾grained variants for pseudo-supervision. A frozen standard-aligned dual encoder provides the global branch, while lightweight adapters and motion extractors lea…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

98 extracted references · 55 canonical work pages

  1. [1]

    2022 , organization=

    Tm2t: Stochastic and tokenized modeling for the reciprocal generation of 3d human motions and texts , author=. 2022 , organization=

    2022

  2. [2]

    Tmr: Text-to-motion retrieval using contrastive 3d human motion synthesis , author=

  3. [3]

    Motiongpt: Human motion as a foreign language , author=

  4. [4]

    A cross-dataset study for text-based 3d human motion retrieval , author=

  5. [5]

    2025 , publisher=

    Motionllm: Understanding human behaviors from human motions and videos , author=. 2025 , publisher=

    2025

  6. [6]

    arXiv preprint arXiv:2410.21747 , year=

    Motiongpt-2: A general-purpose motion-language model for motion generation and understanding , author=. arXiv preprint arXiv:2410.21747 , year=

    arXiv

  7. [7]

    Big data , volume=

    The kit motion-language dataset , author=. Big data , volume=. 2016 , publisher=

    2016

  8. [8]

    Proceedings of the 28th ACM international conference on multimedia , pages=

    Action2motion: Conditioned generation of 3d human motions , author=. Proceedings of the 28th ACM international conference on multimedia , pages=

Show all 98 references
  1. [9]

    BABEL: Bodies, action and behavior with english labels , author=

  2. [10]

    2022 , organization=

    Motionclip: Exposing human motion generation to clip space , author=. 2022 , organization=

    2022

  3. [11]

    Lamp: Language-motion pretraining for motion generation, retrieval, and captioning , author=

  4. [12]

    Avatargpt: All-in-one framework for motion understanding planning generation and beyond , author=

  5. [13]

    Motiongpt3: Human motion as a second modality , author=

  6. [14]

    arXiv preprint arXiv:2410.11404 , year=

    MoChat: Joints-Grouped Spatio-Temporal Grounding LLM for Multi-Turn Motion Comprehension and Description , author=. arXiv preprint arXiv:2410.11404 , year=

    arXiv

  7. [15]

    Fg-t2m: Fine-grained text-driven human motion generation via diffusion model , author=

  8. [16]

    Finemogen: Fine-grained spatio-temporal motion generation and editing , author=

  9. [17]

    2024 , organization=

    Parco: Part-coordinating text-to-motion synthesis , author=. 2024 , organization=

    2024

  10. [18]

    Guided motion diffusion for controllable human motion synthesis , author=

  11. [19]

    2024 , organization=

    Tlcontrol: Trajectory and language control for human motion synthesis , author=. 2024 , organization=

    2024

  12. [20]

    arXiv preprint arXiv:2512.24321 , year=

    UniAct: Unified Motion Generation and Action Streaming for Humanoid Robots , author=. arXiv preprint arXiv:2512.24321 , year=

  13. [21]

    Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , pages=

    Motion generation from fine-grained textual descriptions , author=. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) , pages=

    2024

  14. [22]

    Genmo: A generalist model for human motion , author=

  15. [23]

    arXiv preprint arXiv:2411.19786 , volume=

    Learning motion-text diffusion model for multiple generation tasks , author=. arXiv preprint arXiv:2411.19786 , volume=

    arXiv

  16. [24]

    Advances in Neural Information Processing Systems , volume=

    Snapmogen: Human motion generation from expressive texts , author=. Advances in Neural Information Processing Systems , volume=

  17. [25]

    Dense Motion Captioning , author =

  18. [26]

    Exploring Vision Transformers for 3D Human Motion-Language Models with Motion Patches , author=

  19. [27]

    SGAR: Structural Generative Augmentation for 3D Human Motion Retrieval , author=

  20. [28]

    Kinmo: Kinematic-aware human motion understanding and generation , author=

  21. [29]

    2024 , organization=

    HumanTOMATO: Text-aligned Whole-body Motion Generation , author=. 2024 , organization=

    2024

  22. [30]

    M-adaptor: Text-driven whole-body human motion generation , author=

  23. [31]

    arXiv preprint arXiv:2504.21738 , year=

    Langwbc: Language-directed humanoid whole-body control via end-to-end learning , author=. arXiv preprint arXiv:2504.21738 , year=

    arXiv

  24. [32]

    arXiv preprint arXiv:2412.14172 , year=

    Learning from massive human videos for universal humanoid pose control , author=. arXiv preprint arXiv:2412.14172 , year=

    arXiv

  25. [33]

    Human Motion Diffusion Model , author=

  26. [34]

    5: A real-time controllable vision-language-motion model , author=

    Being-m0. 5: A real-time controllable vision-language-motion model , author=. arXiv preprint arXiv:2508.07863 , year=

    arXiv

  27. [35]

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

  28. [36]

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

  29. [37]

    Pavlakos, Georgios and Choutas, Vasileios and Ghorbani, Nima and Bolkart, Timo and Osman, Ahmed A. A. and Tzionas, Dimitrios and Black, Michael J. , booktitle = CVPR, pages =. Expressive Body Capture:

  30. [38]

    Li, Tianye and Bolkart, Timo and Black, Michael. J. and Li, Hao and Romero, Javier , journal =. Learning a model of facial shape and expression from. 2017 , pages =

    2017

  31. [39]

    Go to zero: Towards zero-shot motion generation with million-scale data , author=

  32. [40]

    Motionstreamer: Streaming motion generation via diffusion-based autoregressive model in causal latent space , author=

  33. [41]

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

  34. [42]

    Mmm: Generative masked motion model , author=

  35. [43]

    Salad: Skeleton-aware latent diffusion for text-driven motion generation and editing , author=

  36. [44]

    Mg-motionllm: A unified framework for motion comprehension and generation across multiple granularities , author=

  37. [45]

    arXiv preprint arXiv:2509.23635 , year=

    Motionverse: A unified multimodal framework for motion comprehension, generation and editing , author=. arXiv preprint arXiv:2509.23635 , year=

  38. [46]

    arXiv preprint arXiv:2411.17335 , year=

    VersatileMotion: A Unified Framework for Motion Synthesis and Comprehension , author=. arXiv preprint arXiv:2411.17335 , year=

    arXiv

  39. [47]

    M3GPT: An Advanced Multimodal, Multitask Framework for Motion Comprehension and Generation , author=

  40. [48]

    Trace and pace: Controllable pedestrian animation via guided trajectory diffusion , author=

  41. [49]

    VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks , author=

  42. [50]

    LoRA: Low-Rank Adaptation of Large Language Models , author=

  43. [51]

    Tri-modal motion retrieval by learning a joint embedding space , author=

  44. [52]

    2024 , organization=

    Chronologically accurate retrieval for temporal grounding of motion-language models , author=. 2024 , organization=

    2024

  45. [53]

    Cross-Modal Quantization for Co-Speech Gesture Generation , year=

    Wang, Zheng and Zhang, Wei and Ye, Long and Zeng, Dan and Mei, Tao , journal=. Cross-Modal Quantization for Co-Speech Gesture Generation , year=

  46. [54]

    EmotionGesture: Audio-Driven Diverse Emotional Co-Speech 3D Gesture Generation , year=

    Qi, Xingqun and Liu, Chen and Li, Lincheng and Hou, Jie and Xin, Haoran and Yu, Xin , journal=. EmotionGesture: Audio-Driven Diverse Emotional Co-Speech 3D Gesture Generation , year=

  47. [55]

    Implicit Compositional Generative Network for Length-Variable Co-Speech Gesture Synthesis , year=

    Xu, Chenghao and Yan, Jiexi and Yang, Yanhua and Deng, Cheng , journal=. Implicit Compositional Generative Network for Length-Variable Co-Speech Gesture Synthesis , year=

  48. [56]

    Total Generate: Cycle in Cycle Generative Adversarial Networks for Generating Human Faces, Hands, Bodies, and Natural Scenes , year=

    Tang, Hao and Sebe, Nicu , journal=. Total Generate: Cycle in Cycle Generative Adversarial Networks for Generating Human Faces, Hands, Bodies, and Natural Scenes , year=

  49. [57]

    Interaction Transformer for Human Reaction Generation , year=

    Chopin, Baptiste and Tang, Hao and Otberdout, Naima and Daoudi, Mohamed and Sebe, Nicu , journal=. Interaction Transformer for Human Reaction Generation , year=

  50. [58]

    Music-Driven Choreography Based on Music Feature Clusters and Dynamic Programming , year=

    Lin, Shuhong and Zukerman, Moshe and Yan, Hong , journal=. Music-Driven Choreography Based on Music Feature Clusters and Dynamic Programming , year=

  51. [59]

    and Geng, Weidong and Li, Xiangdong , journal=

    Sun, Guofei and Wong, Yongkang and Cheng, Zhiyong and Kankanhalli, Mohan S. and Geng, Weidong and Li, Xiangdong , journal=. DeepDance: Music-to-Dance Motion Choreography With Adversarial Learning , year=

  52. [60]

    Murat , journal=

    Ofli, Ferda and Erzin, Engin and Yemez, Yücel and Tekalp, A. Murat , journal=. Learn2Dance: Learning Statistical Music-to-Dance Mappings for Choreography Synthesis , year=

  53. [61]

    Text2Avatar: Articulated 3D Avatar Creation With Text Instructions , year=

    Kwon, Yong-Hoon and Yoon, Ju Hong and Park, Min-Gyu , journal=. Text2Avatar: Articulated 3D Avatar Creation With Text Instructions , year=

  54. [62]

    Learning transferable visual models from natural language supervision , author=

  55. [63]

    arXiv preprint arXiv:1807.03748 , year=

    Representation learning with contrastive predictive coding , author=. arXiv preprint arXiv:1807.03748 , year=

    arXiv

  56. [64]

    MaskControl: Spatio-Temporal Control for Masked Motion Synthesis , author =

  57. [65]

    Motionlcm: Real-time controllable motion generation via latent consistency model , author=

  58. [66]

    Pinxin Liu and Luchuan Song and Junhua Huang and Chenliang Xu , booktitle=ICCV, year=

  59. [67]

    ActFormer: A GAN-based Transformer towards General Action-Conditioned 3D Human Motion Generation , author=

  60. [68]

    ReGenNet: Towards Human Action-Reaction Synthesis , author=

  61. [69]

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

  62. [70]

    arXiv preprint arXiv:2510.26794 , year=

    The quest for generalizable motion generation: Data, model, and evaluation , author=. arXiv preprint arXiv:2510.26794 , year=

  63. [71]

    arXiv preprint arXiv:2512.23464 , year=

    HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generation , author=. arXiv preprint arXiv:2512.23464 , year=

  64. [72]

    arXiv preprint arXiv:2307.10894 , year=

    Human motion generation: A survey , author=. arXiv preprint arXiv:2307.10894 , year=

    arXiv

  65. [73]

    arXiv preprint arXiv:2511.07820 , year=

    Sonic: Supersizing motion tracking for natural humanoid whole-body control , author=. arXiv preprint arXiv:2511.07820 , year=

    arXiv

  66. [74]

    Inter-x: Towards versatile human-human interaction analysis , author=

  67. [75]

    Perceiving and Acting in First-Person: A Dataset and Benchmark for Egocentric Human-Object-Human Interactions , author=

  68. [76]

    arXiv preprint arXiv:2111.07783 , year=

    Filip: Fine-grained interactive language-image pre-training , author=. arXiv preprint arXiv:2111.07783 , year=

    arXiv

  69. [77]

    Pyramidclip: Hierarchical feature alignment for vision-language model pretraining , author=

  70. [78]

    Fine-grained video-text retrieval with hierarchical graph reasoning , author=

  71. [79]

    Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

    Benchmark Granularity and Model Robustness for Image-Text Retrieval: A Reproducibility Study , author=. Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

  72. [80]

    Proceedings of the 46th international ACM SIGIR conference on research and development in information retrieval , pages=

    Rethinking benchmarks for cross-modal image-text retrieval , author=. Proceedings of the 46th international ACM SIGIR conference on research and development in information retrieval , pages=

  73. [81]

    Coot: Cooperative hierarchical transformer for video-text representation learning , author=

  74. [82]

    Hit: Hierarchical transformer with momentum contrast for video-text retrieval , author=

  75. [83]

    IEEE Access , year=

    Tencent text-video retrieval: hierarchical cross-modal interactions with multi-level representations , author=. IEEE Access , year=

  76. [84]

    On the continuity of rotation representations in neural networks , author=

  77. [85]

    Motion-x: A large-scale 3d expressive whole-body human motion dataset , author=

  78. [86]

    ACM Transactions on Graphics (TOG) , volume=

    MotionBricks: Scalable Real-Time Motions with Modular Latent Generative Model and Smart Primitives , author=. ACM Transactions on Graphics (TOG) , volume=. 2026 , publisher=

    2026

  79. [87]

    2025 , organization=

    ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model , author=. 2025 , organization=

    2025

  80. [88]

    arXiv preprint arXiv:2501.05098 , year=

    Motion-x++: A large-scale multimodal 3d whole-body human motion dataset , author=. arXiv preprint arXiv:2501.05098 , year=

    arXiv

  81. [89]

    2026 , month = apr, url =

    2026

  82. [90]

    2024 , eprint=

    Making Text Embedders Few-Shot Learners , author=. 2024 , eprint=

    2024

  83. [91]

    2025 , eprint=

    Qwen2.5 Technical Report , author=. 2025 , eprint=

    2025

  84. [92]

    2025 , month = nov, note =

    2025

  85. [93]

    and Varol, G

    Petrovich, Mathis and Black, Michael J. and Varol, G. European Conference on Computer Vision (

  86. [94]

    and Pons-Moll, Gerard and Black, Michael J

    Mahmood, Naureen and Ghorbani, Nima and Troje, Nikolaus F. and Pons-Moll, Gerard and Black, Michael J. , booktitle =. 2019 , month_numeric =

    2019

  87. [95]

    European conference on computer vision , pages=

    Long-clip: Unlocking the long-text capability of clip , author=. European conference on computer vision , pages=. 2024 , organization=

    2024

  88. [96]

    Llamo: Scaling pretrained language models for unified motion understanding and generation with continuous autoregressive tokens , author=

  89. [97]

    2020 , organization=

    A simple framework for contrastive learning of visual representations , author=. 2020 , organization=

    2020

  90. [98]

    Momentum contrast for unsupervised visual representation learning , author=

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.