mlsanityData to 14 Oct 2026

ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model

arXiv 2412.14559 ↗First posted 19 Dec 2024cs.CV cs.LGcited by 53

Not on the map. Why?

Abstract (arXiv)

The scaling law has been validated in various domains, such as natural language processing (NLP) and massive computer vision tasks; however, its application to motion generation remains largely unexplored. In this paper, we introduce a scalable motion generation framework that includes the motion tokenizer Motion FSQ-VAE and a text-prefix autoregressive transformer. Through comprehensive experiments, we observe the scaling behavior of this system. For the first time, we confirm the existence of scaling laws within the context of motion generation. Specifically, our results demonstrate that the normalized test loss of our prefix autoregressive models adheres to a logarithmic law in relation to compute budgets. Furthermore, we also confirm the power law between Non-Vocabulary Parameters, Vocabulary Parameters, and Data Tokens with respect to compute budgets respectively. Leveraging the scaling law, we predict the optimal transformer size, vocabulary size, and data requirements for a compute budget of 1e18. The test loss of the system, when trained with the optimal model size, vocabulary size, and required data, aligns precisely with the predicted test loss, thereby validating the scaling law.

Key facts. Cited by 53 papers in the map (data through 14 Oct 2026). Builds on 65 of its 76 references that are in the map. First posted on arXiv 19 Dec 2024.

Builds on

65 of 76 references.

References · 65

  1. 2017Attention Is All You Needcited by 60,361
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  3. 2020Language Models are Few-Shot Learnerscited by 32,419
  4. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  5. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,863
  6. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  7. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  8. 2020Denoising Diffusion Implicit Modelscited by 9,781
  9. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  10. 2020Scaling Laws for Neural Language Modelscited by 6,586
  11. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,585
  12. 2023Qwen Technical Reportcited by 3,344
38 more
  1. 2022Training Compute-Optimal Large Language Modelscited by 3,019
  2. 2019AMASS: Archive of Motion Capture as Surface Shapescited by 1,236
  3. 2022Human Motion Diffusion Modelcited by 1,080
  4. 2024Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Predictioncited by 1,023
  5. 2022MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Modelcited by 780
  6. 2024Autoregressive Model Beats Diffusion: Llama for Scalable Image Generationcited by 777
  7. 2024DeepSeek LLM: Scaling Open-Source Language Models with Longtermismcited by 642
  8. 2023T2M-GPT: Generating Human Motion from Textual Descriptions with Discrete Representationscited by 588
  9. 2017Deep Learning Scaling is Predictable, Empiricallycited by 582
  10. 2022Executing your Commands via Motion Diffusion in Latent Spacecited by 575
  11. 2023MotionGPT: Human Motion as a Foreign Languagecited by 527
  12. 2022TEMOS: Generating diverse human motions from textual descriptionscited by 492
  13. 2020Action2Motion: Conditioned Generation of 3D Human Motionscited by 441
  14. 2023Finite Scalar Quantization: VQ-VAE Made Simplecited by 440
  15. 2022MotionCLIP: Exposing Human Motion Generation to CLIP Spacecited by 408
  16. 2023MoMask: Generative Masked Modeling of 3D Human Motionscited by 400
  17. 2022PhysDiff: Physics-Guided Human Motion Diffusion Modelcited by 368
  18. 2022TM2T: Stochastic and Tokenized Modeling for the Reciprocal Generation of 3D Human Motions and Textscited by 359
  19. 2023Human Motion Diffusion as a Generative Priorcited by 357
  20. 2016The KIT Motion-Language Datasetcited by 338
  21. 2023ReMoDiffuse: Retrieval-Augmented Motion Diffusion Modelcited by 287
  22. 2019Language2Pose: Natural Language Grounded Pose Forecastingcited by 257
  23. 2023OmniControl: Control Any Joint at Any Time for Human Motion Generationcited by 238
  24. 2023Motion-X: A Large-scale 3D Expressive Whole-body Human Motion Datasetcited by 227
  25. 2023Guided Motion Diffusion for Controllable Human Motion Synthesiscited by 223
  26. 2022MoFusion: A Framework for Denoising-Diffusion-based Motion Synthesiscited by 218
  27. 2022TEACH: Temporal Action Composition for 3D Humanscited by 176
  28. 2024MotionLCM: Real-time Controllable Motion Generation via Latent Consistency Modelcited by 168
  29. 2022HUMANISE: Language-conditioned Human Motion Generation in 3D Scenescited by 158
  30. 2023MotionGPT: Finetuned LLMs Are General-Purpose Motion Generatorscited by 145
  31. 2024Scaling Laws with Vocabulary: Larger Models Deserve Larger Vocabulariescited by 118
  32. 2017Text2Action: Generative Adversarial Synthesis from Language to Actioncited by 106
  33. 2021Text2Gestures: A Transformer-Based Network for Generating Emotive Body Gestures for Virtual Agentscited by 106
  34. 2023HumanTOMATO: Text-aligned Whole-body Motion Generationcited by 104
  35. 2023Inter-X: Towards Versatile Human-Human Interaction Analysiscited by 98
  36. 2023Unified Human-Scene Interaction via Prompted Chain-of-Contactscited by 94
  37. 2023TLControl: Trajectory and Language Control for Human Motion Synthesiscited by 94
  38. 2024Multi-Track Timeline Control for Text-Driven 3D Human Motion Generationcited by 91

50 most cited of 65 All 65 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 21; 2026: 32.

53 papers.

Sort:

By area:

  1. 2025MotionStreamer: Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Spacecited by 86
  2. 2025Go to Zero: Towards Zero-shot Motion Generation with Million-scale Datacited by 68
  3. 2025TokenHSI: Unified Synthesis of Physical Human-Scene Interactions through Task Tokenizationcited by 51
  4. 2025SnapMoGen: Human Motion Generation from Expressive Textscited by 47
  5. 2026Kimodo: Scaling Controllable Human Motion Generationcited by 37
  6. 2025HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generationcited by 36
  7. 2025Absolute Coordinates Make Motion Generation Easycited by 29
  8. 2025The Quest for Generalizable Motion Generation: Data, Model, and Evaluationcited by 22
  9. 2025Simulating the Real World: A Unified Survey of Multimodal Generative Modelscited by 15
  10. 2025CoDA: Coordinated Diffusion Noise Optimization for Whole-Body Manipulation of Articulated Objectscited by 12
  11. 2025MOSPA: Human Motion Generation Driven by Spatial Audiocited by 12
  12. 2025Text2Interact: High-Fidelity and Diverse Text-to-Two-Person Interaction Generationcited by 11
38 more
  1. 2026LLaMo: Scaling Pretrained Language Models for Unified Motion Understanding and Generation with Continuous Autoregressive Tokenscited by 11
  2. 2025Being-M0.5: A Real-Time Controllable Vision-Language-Motion Modelcited by 10
  3. 2025EgoReAct: Egocentric Video-Driven 3D Human Reaction Generationcited by 7
  4. 2025HumanMM: Global Human Motion Recovery from Multi-shot Videoscited by 6
  5. 2025IRG-MotionLLM: Interleaving Motion Generation, Assessment and Refinement for Text-to-Motion Generationcited by 6
  6. 2026CoMoVi: Co-Generation of 3D Human Motions and Realistic Videoscited by 6
  7. 2026ActionPlan: Future-Aware Streaming Motion Synthesis via Frame-Level Action Planningcited by 5
  8. 2026UMO: Unified In-Context Learning Unlocks Motion Foundation Model Priorscited by 5
  9. 2025X-MoGen: Unified Motion Generation across Humans and Animalscited by 4
  10. 2025VimoRAG: Video-based Retrieval-augmented 3D Motion Generation for Motion Language Modelscited by 4
  11. 2025OmniMotion-X: Versatile Multimodal Whole-Body Motion Generationcited by 4
  12. 2026Bridging Semantic and Kinematic Conditions with Diffusion-based Discrete Motion Tokenizercited by 4
  13. 2026Next-Scale Autoregressive Models for Text-to-Motion Generationcited by 4
  14. 2026MotionHiFlow: Text-to-motion via hierarchical flow matchingcited by 4
  15. 2026ScaleMoGen: Autoregressive Next-Scale Prediction for Human Motion Generationcited by 4
  16. 2026PRISM: Streaming Human Motion Generation with Per-Joint Latent Decompositioncited by 3
  17. 2025MOGO: Residual Quantized Hierarchical Causal Transformer for High-Quality and Real-Time 3D Human Motion Generationcited by 2
  18. 2025Topology-Agnostic Animal Motion Generation from Text Promptcited by 2
  19. 2026Language-Guided Transformer Tokenizer for Human Motion Generationcited by 2
  20. 2026U-Mind: A Unified Framework for Real-Time Multimodal Interaction with Audiovisual Generationcited by 2
  21. 2026Riemannian Motion Generation: A Unified Framework for Human Motion Representation and Generation via Riemannian Flow Matchingcited by 2
  22. 2026OpenT2M: No-frill Motion Generation with Open-source,Large-scale, High-quality Datacited by 2
  23. 2026HandX: Scaling Bimanual Motion and Interaction Generationcited by 2
  24. 2026Seeing Without Eyes: 4D Human-Scene Understanding from Wearable IMUscited by 2
  25. 2026ReactMotion: Generating Reactive Listener Motions from Speaker Utterancecited by 1
  26. 2026IAM: Identity-Aware Human Motion and Shape Joint Generationcited by 1
  27. 2026UMo: Unified Sparse Motion Modeling for Real-Time Co-Speech Avatarscited by 1
  28. 2026AnyMo: Scaling Any-Modality Conditional Motion Generation with Masked Modelingcited by 1
  29. 2026Beyond MoCap: Scaling Motion Tokenizers with Synthetic Human Motion for Generative Modelingcited by 1
  30. 2026ReactiveBFM: Reactive Closed-Loop Motion Planning Towards Universal Humanoid Whole-Body Controlcited by 1
  31. 2026MRBench: A Comprehensive Benchmark for Human Motion-Text Retrievalcited by 1
  32. 2026MoSAT: Human Motion Generation from Spatial Audio and Textual Descriptioncited by 1
  33. 2025MoGIC: Boosting Motion Generation via Intention Understanding and Visual Contextcited by 0
  34. 2026Marrying Text-to-Motion Generation with Skeleton-Based Action Recognitioncited by 0
  35. 2026Multi-scale Coarse-to-fine Modeling for Test-time Human Motion Controlcited by 0
  36. 2026MotionMERGE: A Multi-granular Framework for Human Motion Editing, Reasoning, Generation, and Explanationcited by 0
  37. 2026MUGEN: A Unified Framework for Efficient Motion Understanding and Generationcited by 0
  38. 2026GenPrior: Unleashing Text-to-Motion Generative Priors for Zero-Shot Skeleton-based Action Recognitioncited by 0

50 most cited of 53 All 53 (JSON)