Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

TSTMotion: Training-free Scene-aware Text-to-motion Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Text-to-motion can be made scene-aware with zero training by steering a diffusion model with LLM-planned keyframes.

desk verdict A genuinely training-free scene-aware text-to-motion pipeline worth engaging with, but the quantitative evidence is partly circular and the LLM Motion Planner — the linchpin — is never validated in isolation. read the letter →

arxiv 2505.01182 v2 pith:VHUZN7UQ submitted 2025-05-02 cs.CV cs.AI

Ziyan Guo , Haoxuan Qu , Hossein Rahmani , Dewen Soh , Ping Hu , Qiuhong Ke , Jun Liu This is my paper · ORCID
classification cs.CVcs.AI
keywords training-freescene-awaretext-to-motionmotiondiffusionmodelsLLMplanningsigneddistancefieldfoundationhuman-sceneinteractiongradientconditioning
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 introduces TSTMotion, the first training-free framework for scene-aware text-to-motion generation. It claims that a pre-trained blank-background text-to-motion diffusion model can be made scene-aware at inference time by conditioning it on a motion guidance -- a masked skeleton sequence of keyframe joint positions -- produced by an LLM from a scene-derived roadmap and height map. Two training-free gradient modifications in the reverse diffusion process enforce alignment with this guidance and penalize penetration into the scene mesh, so no scene-aware motion dataset and no model weight updates are needed. If this works, scene-aware motion generation becomes cheap and generalizes to new indoor and outdoor scenes, and the paper reports superior or competitive results against training-based baselines on HUMANISE, AffordMotion, and PROX.

What carries the argument

The central object is the motion guidance $s[M_{\text{mask}}]$: a sequence of 3D joint positions over a few keyframes, masked to the joints that matter for the specified text–scene interaction. The mechanism that carries the argument is the pairing of this guidance with two training-free gradient modifications applied during the reverse diffusion of the motion model: an alignment loss (Eq. 5–6) between the diffuser's predicted clean motion and the guidance, and a contact/penetration penalty (Eq. 7–8) built from a signed distance field of the scene mesh over the skinned SMPL body. The guidance itself comes from an LLM Motion Planner that reads the road map and the height map, so the whole pipeline depends on the LLM's capacity to turn 2D layout into plausible 3D joint placement.

What would settle it

Take a fixed 3D scene, run the Scene Compiler, and ask the Motion Planner to output keyframe pelvis positions for 'sitting on the couch'; compare those positions to the couch's true location. If the planner's sitting keyframes land more than 0.5 meters from the couch in a majority of repeated trials, the downstream alignment gradient would faithfully produce motions to the wrong place, contradicting the central claim.

Watch

Extended reading notes

Core claim

TSTMotion's central claim is that scene awareness can be injected into blank-background motion diffusion models purely at inference time. The framework compiles a 3D scene into a 2D road map of walkable areas and a height map of the target object, asks an LLM (the Motion Planner) to output a sparse skeleton guidance $s[M_{\text{mask}}]$ -- joint positions for the joints and frames relevant to the interaction -- and then steers the diffusion sampler with two gradients: one that aligns the model's predicted clean motion (mapped by forward kinematics) to $s[M_{\text{mask}}]$ (Eq. 6), and one that pushes the SMPL mesh of the body out of the scene's signed distance field (Eq. 8). The authors argue that foundation models already encode enough knowledge about scenes, skeletons, and motion semantics to supply this guidance, so the pipeline avoids specialized data and training. They support the claim with quantitative comparisons on HUMANISE and AffordMotion and qualitative results in PROX and unseen outdoor scenes.

Load-bearing premise

The load-bearing premise is that an LLM, supplied only with a 2D road map and a height map of one target object, can produce a geometrically plausible 3D skeleton keyframe path that correctly locates the object and encodes the requested interaction.

Editorial extensions

If this is right

  • Scene-aware text-to-motion generation no longer requires collecting paired text–motion–scene datasets, which lowers the cost for embodied AI, games, and film.
  • Any pre-trained blank-background motion generator can be repurposed for a new 3D scene by recomputing the spatial auxiliary and re-running the planner, so action diversity scales with the LLM's knowledge rather than dataset annotations.
  • The gradient-alignment trick (Eq. 6) offers a general recipe for conditioning a motion diffusion model on partial skeleton observations at inference time, applicable beyond scene interaction.
  • The SDF-based penetration penalty (Eq. 8) is a plug-and-play collision-avoidance module that can be grafted onto other skinned motion generators.

Reading between the lines

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

  • The Motion Planner is the likely bottleneck: if the LLM mislocates the target or chooses the wrong contact joints, the alignment gradient faithfully generates motion to the wrong place, and the Motion Checker can only filter or restart, not correct the error.
  • A testable extension is to sample multiple planner proposals per scene and let the Motion Checker select the best one, which could raise success rates on complex interactions without any training.
  • Since the alignment gradient is added at every denoising step with fixed strength, a schedule that anneals $\lambda$ over the reverse process might improve stability for long motions.
  • The Scene Compiler's road map + height map encoding might be reusable as a general way to feed 3D scenes to any LLM-based planner, not just this motion pipeline.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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. The paper proposes TSTMotion, a training-free framework for scene-aware text-to-motion generation. It composes a Scene Compiler that converts a 3D scene into a road map and a height map, an LLM-based Motion Planner that produces sparse 3D skeleton keyframes as motion guidance, a pre-trained motion diffusion model modified at inference time by two gradients (Eq. 6 aligns the predicted clean motion to the guidance; Eq. 8 penalizes penetration into the scene mesh), and an LLM-based Motion Checker that restarts the pipeline when the output fails scene/text criteria. Experiments on HUMANISE and AffordMotion, plus qualitative results on PROX, are used to claim superior performance over training-based baselines and generalization to outdoor scenes.

Significance. If fully validated, the framework would be practically attractive because it avoids scene-motion training data and model weight updates, while leveraging pre-trained blank-background motion diffusion models. The paper has clear strengths: it releases code/project page, decomposes the problem into interpretable foundation-model components, and provides ablations showing each component contributes on the chosen metrics. The core idea of conditioning a pre-trained MDM via posterior-style gradients on an LLM-generated skeleton guidance is coherent and worth pursuing. However, the current validation is insufficient in two load-bearing respects: the central LLM planning step is not evaluated in isolation, and several headline metrics largely restate the optimization objectives rather than providing independent evidence of scene-awareness.

major comments (4)
  1. [Section II-B and Fig.2] The Motion Planner is the linchpin of the training-free claim: it must convert a 2D road map, a height map of the target, and a text prompt into absolute 3D joint keyframes that correctly localize the target and encode the interaction. The manuscript provides no isolated evaluation of this conversion—no success rate for the planner, no Motion Checker acceptance statistics, and no failure-case analysis. Since Eq. (6) faithfully pulls the generated motion toward s[Mmask] and the Motion Checker (Sec. II-D) can only restart the same planner, a plan with a mislocated target or an unreachable pelvis position cannot be corrected downstream. I therefore cannot yet accept the claim of superior performance over training-based methods; the paper should add an oracle-style ablation (e.g., replacing the LLM plan with ground-truth contact/goal keyframes) and report planner/checker success rates.
  2. [Section II-B and Eq. (5)] The construction of the dense guidance tensor s[Mmask] is underspecified. The LLM output in Fig.2 is a small set of sparse keyframes (keyframe_1, keyframe_20, keyframe_40) with only a 'pelvis' joint, whereas Eq. (5) requires a skeleton sequence in R^{N x J x 3} over all N frames and masked joints. No interpolation, forward-kinematic initialization, or temporal filling procedure is described. Without this step, the alignment loss in Eq. (5) cannot be evaluated, so the main mechanism of the method is not fully reproducible from the text. Please specify the exact conversion from the LLM JSON to s[Mmask].
  3. [Section III-A and Section II-C] Three of the headline metrics are near-verbatim restatements of the two inference-time losses. Body-to-goal distance and contact score measure the same proximity to the target that Eq. (6) is explicitly minimizing, and non-collision score measures the same penetration avoidance as Eq. (8). Consequently, the improvements in Tables I and III are partly guaranteed by construction and do not constitute independent evidence that the generated motions are scene-appropriate. The independent evidence (FID, R-Precision, Multimodal Distance) is reported only for the AffordMotion benchmark; the paper should also report those metrics on HUMANISE, or otherwise separate the optimization objective from the evaluation.
  4. [Section II-D and Table III] The claim that 'only one iteration is necessary' for the Motion Checker is not supported by any experiment varying the iteration count; the improvement from w/o Motion Checker to Ours in Table III is also small (e.g., contact 0.89 vs 0.91). Since the checker is presented as a core component and a restart mechanism, the paper should either show a saturation curve over iterations or soften the claim.
minor comments (5)
  1. [Eq. (6)] The gradient is written as ∇_{x_k} but the loss is evaluated on x̂_k^0; please clarify whether the update is w.r.t. x_k through the denoiser map or directly on x̂_k^0, and state the gradient-stop convention.
  2. [Section III-D] The text says 'As shown in Table I' when discussing ablation variants; the results are actually in Table III.
  3. [Section III-B] Only single values for λ=2 and η=0.5 are given; since these hyperparameters control the two training-free modifications, a sensitivity analysis would strengthen the reproducibility of the claims.
  4. [Section III-A] Quality and action scores are described as perceptual studies, but no details on participant number, instructions, or consistency are provided, which makes the significance tests in Table I hard to interpret.
  5. [General] There are minor typos: 'Mode,l' in the Section II overview paragraph and 'Grountruth' in Table II should be corrected.

Circularity Check

2 steps flagged · score 6.0 of 10

Evaluation metrics are near-verbatim proxies for the inference-time losses, so high scene-interaction scores are partly self-verifying; the training-free pipeline itself is not circular.

  1. other [Sec. II-C, Eq. 7-8, versus Sec. III-A, 'Evaluation metrics']
    "Lscene(ˆxk0,P ) = ReLU (−SDF (SMPL (ˆxk0),P )) ... the non-collision score to evaluate the non-overlapping degree between the human and objects in scenes"

    The non-collision metric is defined as non-overlap between human and scene, which is exactly what Lscene computes as penetration depth (ReLU of negative SDF). Equation 8 then explicitly subtracts the gradient of Lscene at every reverse step, so the reported high non-collision score is the optimizer reporting its own objective rather than independent evidence of scene awareness. Contact score, defined as body-to-scene distance under a threshold, is likewise directly shaped by Eq. 6, which aligns masked joints to the LLM's contact keyframes; body-to-goal distance is tied to the goal keyframe in s[Mmask]. These metrics are therefore not independent of the method's construction.

  2. other [Sec. II-D, Motion Checker]
    "it verifies whether the motion is beyond the scene and matches the motion semantics. If the generated motion sequences fail to meet these criteria, the framework initiates a restart from the Motion Planner."

    The final motion is obtained only after outputs that fail the checker's scene and semantics criteria are rejected and regenerated. Subsequent contact, non-collision, and action scores are therefore measured on a filtered, self-selected subset of outputs that were already required to satisfy the same qualitative conditions. This makes high values on these metrics partly guaranteed by the selection rule rather than by unbiased generation, weakening the claim that the superiority is an independent empirical outcome.

full rationale

The methodological derivation chain is not circular in the strong sense: the Scene Compiler converts a point cloud into a road map and height map, the Motion Planner LLM outputs a sparse skeleton s[Mmask], and the Aligned Motion Diffusion Model is steered by two gradient losses (Eq. 6 and Eq. 8). No parameter is fitted to the evaluation metrics, and no load-bearing self-citation or imported uniqueness theorem forces the central design. The main circularity is in the validation. Non-collision score is semantically identical to Lscene in Eq. 7, which Eq. 8 directly minimizes; contact score is the body-to-scene distance produced by aligning masked joints to LLM-chosen contact keyframes; body-to-goal distance is tied to the goal keyframe in s[Mmask]. The Motion Checker further restarts any motion that fails its scene/semantics criteria, so final outputs are selected on the same dimensions later scored. Thus high contact, non-collision, and body-to-goal values are partly guaranteed by construction and do not independently establish scene-aware quality. Meanwhile, the LLM planner's ability to infer correct 3D keyframes from a 2D road map and a target height map is not validated in isolation; that is a correctness risk rather than a circularity. Reference [16] is a minor self-citation used to justify iterative refinement, but it is not load-bearing for the central claim. Overall, the framework has independent content, but the reported scene-interaction metrics are partly self-verifying, warranting a partial circularity score of 6.

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

The central claim depends on four hand-set hyperparameters and five background assumptions about foundation models. The largest burden is that unvalidated LLM reasoning and segmentation outputs are the sole source of the scene-aware guidance; the paper does not measure intermediate error. These are engineering assumptions rather than physical axioms, but they determine whether the training-free formula actually works.

free parameters (4)
  • lambda (alignment strength) = 2
    Hand-set in Section III-B for Eq. 6; no sensitivity analysis or validation split is reported.
  • eta (overlap penalty strength) = 0.5
    Hand-set in Section III-B for Eq. 8; no sensitivity analysis.
  • M (number of rendered views) = 16
    Set in Section III-B following Huang et al. [18]; affects object recognition quality.
  • Motion Checker iteration count = 1
    Fixed in Section III-B; the paper says one iteration is sufficient but provides no study of how restarts affect metrics.
assumptions (5)
  • domain assumption LLMs can infer geometrically plausible 3D skeleton contact positions from a 2D road map, a target height map, and a text prompt.
    Invoked in Section II-B as the basis of the Motion Planner; not validated independently.
  • domain assumption RAM image tagging and OpenIns3D segmentation correctly identify and isolate the target object in the 3D scene.
    Section II-A steps (1)-(2) rely on pretrained taggers/segmenters with no accuracy check on the test scenes.
  • ad hoc to paper The road map and height map representation preserves all scene information needed for motion planning, so texture/appearance can be discarded.
    Stated in Section II-A as a design choice; no experiment compares against richer scene representations.
  • domain assumption The pre-trained motion diffusion model (Xie et al.) provides a motion prior strong enough that per-step gradient nudges stay on manifold.
    Section II-C relies on posterior sampling; no analysis of distribution shift caused by the two gradient modifications.
  • standard math The SMPL skinning and scene mesh SDF give an accurate enough collision signal to reduce overlap without distorting motion.
    Eq. 7 uses standard computational geometry; accuracy depends on mesh quality, which is not analyzed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TSTMotion: Training-free Scene-aware Text-to-motion Generation." pith.science (2026). https://pith.science/paper/VHUZN7UQ

@misc{pith2026250501182,
  author       = {Pith},
  title        = {Pith review of: TSTMotion: Training-free Scene-aware Text-to-motion Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VHUZN7UQ}},
  note         = {Machine review of arXiv:2505.01182}
}
read the original abstract

Text-to-motion generation has recently garnered significant research interest, primarily focusing on generating human motion sequences in blank backgrounds. However, human motions commonly occur within diverse 3D scenes, which has prompted exploration into scene-aware text-to-motion generation methods. Yet, existing scene-aware methods often rely on large-scale ground-truth motion sequences in diverse 3D scenes, which poses practical challenges due to the expensive cost. To mitigate this challenge, we are the first to propose a \textbf{T}raining-free \textbf{S}cene-aware \textbf{T}ext-to-\textbf{Motion} framework, dubbed as \textbf{TSTMotion}, that efficiently empowers pre-trained blank-background motion generators with the scene-aware capability. Specifically, conditioned on the given 3D scene and text description, we adopt foundation models together to reason, predict and validate a scene-aware motion guidance. Then, the motion guidance is incorporated into the blank-background motion generators with two modifications, resulting in scene-aware text-driven motion sequences. Extensive experiments demonstrate the efficacy and generalizability of our proposed framework. We release our code in \href{https://tstmotion.github.io/}{Project Page}.

Figures

Figures reproduced from arXiv: 2505.01182 by the authors.

Figure 1
Figure 1. Illustration of scene-aware text-driven motion sequences generated by [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of our proposed training-free TSTMotion framework for the given text [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of between Wang et al. [3], DIMOS*, AffordMotion and our TSTMotion on the unseen PROX dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. MotionLab: Unified Human Motion Generation and Editing via the Motion-Condition-Motion Paradigm

    cs.CV 2025-02 conditional novelty 6.0 of 10

    MotionLab unifies text-based and trajectory-based motion generation with text-based editing, trajectory-based editing, motion in-betweening, and style transfer in one flow-based transformer.

Reference graph

Works this paper leans on

21 extracted references · 14 canonical work pages · cited by 1 Pith paper

  1. [1]

    Human motion diffusion model,

    Guy Tevet, Sigal Raab, Brian Gordon, Yonatan Shafir, Daniel Cohen-Or, and Amit H Bermano, “Human motion diffusion model,” arXiv preprint arXiv:2209.14916, 2022

    arXiv 2022

  2. [2]

    Motiondiffuse: Text-driven human mo- tion generation with diffusion model,

    Mingyuan Zhang, Zhongang Cai, Liang Pan, Fangzhou Hong, Xinying Guo, Lei Yang, and Ziwei Liu, “Motiondiffuse: Text-driven human mo- tion generation with diffusion model,” arXiv preprint arXiv:2208.15001, 2022

    arXiv 2022

  3. [3]

    Humanise: Language-conditioned human motion generation in 3d scenes,

    Zan Wang, Yixin Chen, Tengyu Liu, Yixin Zhu, Wei Liang, and Siyuan Huang, “Humanise: Language-conditioned human motion generation in 3d scenes,” NeurIPS, vol. 35, pp. 14959–14971, 2022

    work page 2022

  4. [4]

    Unified human-scene interaction via prompted chain-of-contacts,

    Zeqi Xiao, Tai Wang, Jingbo Wang, Jinkun Cao, Wenwei Zhang, Bo Dai, Dahua Lin, and Jiangmiao Pang, “Unified human-scene interaction via prompted chain-of-contacts,” in ICLR, 2024

    work page 2024

  5. [5]

    Scaling up dynamic human-scene interaction modeling,

    Nan Jiang, Zhiyuan Zhang, Hongjie Li, Xiaoxuan Ma, Zan Wang, Yixin Chen, Tengyu Liu, Yixin Zhu, and Siyuan Huang, “Scaling up dynamic human-scene interaction modeling,” in CVPR, 2024, pp. 1737–1747

    work page 2024

  6. [6]

    Move as You Say, Interact as You Can: Language-guided Human Motion Generation with Scene Affordance

    Zan Wang, Yixin Chen, Baoxiong Jia, Puhao Li, Jinlu Zhang, Jingze Zhang, Tengyu Liu, Yixin Zhu, Wei Liang, and Siyuan Huang, “Move as you say, interact as you can: Language-guided human motion generation with scene affordance,” arXiv preprint arXiv:2403.18036 , 2024

    work page Pith review arXiv 2024

  7. [7]

    Stochastic scene-aware motion prediction,

    Mohamed Hassan, Duygu Ceylan, Ruben Villegas, Jun Saito, Jimei Yang, Yi Zhou, and Michael J. Black, “Stochastic scene-aware motion prediction,” in ICCV, October 2021, pp. 11374–11384

    work page 2021

  8. [8]

    Synthesizing Diverse Human Motions in 3D Indoor Scenes

    Kaifeng Zhao, Yan Zhang, Shaofei Wang, Thabo Beeler, and Siyu Tang, “Synthesizing diverse human motions in 3d indoor scenes,” arXiv preprint arXiv:2305.12411, 2023

    work page Pith review arXiv 2023

Show all 21 references
  1. [9]

    Generating diverse and natural 3d human motions from text,

    Chuan Guo, Shihao Zou, Xinxin Zuo, Sen Wang, Wei Ji, Xingyu Li, and Li Cheng, “Generating diverse and natural 3d human motions from text,” in CVPR, 2022, pp. 5152–5161

    2022

  2. [10]

    Chain-of-thought prompting elicits reasoning in large language models,

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al., “Chain-of-thought prompting elicits reasoning in large language models,”NeurIPS, vol. 35, pp. 24824– 24837, 2022

    2022

  3. [11]

    Visual language maps for robot navigation,

    Chenguang Huang, Oier Mees, Andy Zeng, and Wolfram Burgard, “Visual language maps for robot navigation,” in ICRA. IEEE, 2023, pp. 10608–10615

    2023

  4. [12]

    Tabllm: Few-shot classification of tabular data with large language models,

    Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David Sontag, “Tabllm: Few-shot classification of tabular data with large language models,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2023, pp. 5549–5581

    2023

  5. [13]

    Language models are few-shot learners,

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al., “Language models are few-shot learners,” NeurIPS, vol. 33, pp. 1877–1901, 2020

    1901

  6. [14]

    Diffusion posterior sampling for general noisy inverse problems,

    Hyungjin Chung, Jeongsol Kim, Michael T Mccann, Marc L Klasky, and Jong Chul Ye, “Diffusion posterior sampling for general noisy inverse problems,” arXiv preprint arXiv:2209.14687 , 2022

    2022 arXiv

  7. [15]

    Ex- pressive body capture: 3D hands, face, and body from a single image,

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black, “Ex- pressive body capture: 3D hands, face, and body from a single image,” in CVPR, 2019, pp. 10975–10985

    2019

  8. [16]

    Enhancing human-centered dynamic scene understanding via multiple llms collab- orated reasoning,

    Hang Zhang, Wenxiao Zhang, Haoxuan Qu, and Jun Liu, “Enhancing human-centered dynamic scene understanding via multiple llms collab- orated reasoning,” arXiv preprint arXiv:2403.10107 , 2024

    2024 arXiv

  9. [17]

    Recognize anything: A strong image tagging model,

    Youcai Zhang, Xinyu Huang, Jinyu Ma, Zhaoyang Li, Zhaochuan Luo, Yanchun Xie, Yuzhuo Qin, Tong Luo, Yaqian Li, Shilong Liu, et al., “Recognize anything: A strong image tagging model,” in CVPR, 2024, pp. 1724–1732

    2024

  10. [18]

    Openins3d: Snap and lookup for 3d open-vocabulary instance segmentation,

    Zhening Huang, Xiaoyang Wu, Xi Chen, Hengshuang Zhao, Lei Zhu, and Joan Lasenby, “Openins3d: Snap and lookup for 3d open-vocabulary instance segmentation,” arXiv preprint arXiv:2309.00616 , 2023

    2023 arXiv

  11. [19]

    Gpt-4 technical report,

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al., “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  12. [20]

    Omnicontrol: Control any joint at any time for human motion generation,

    Yiming Xie, Varun Jampani, Lei Zhong, Deqing Sun, and Huaizu Jiang, “Omnicontrol: Control any joint at any time for human motion generation,” arXiv preprint arXiv:2310.08580 , 2023

    2023 arXiv

  13. [21]

    Resolving 3D human pose ambiguities with 3D scene con- straints,

    Mohamed Hassan, Vasileios Choutas, Dimitrios Tzionas, and Michael J. Black, “Resolving 3D human pose ambiguities with 3D scene con- straints,” in International Conference on Computer Vision , Oct. 2019, pp. 2282–2292

    2019

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.