Research map / Video / Video generation
Think Before You Move: Latent Motion Reasoning for Text-to-Motion Generation
Yijie Qian, Juncheng Wang, Yuxiang Feng, Chao Xu, Wang Lu, Yang Liu, Baigui Sun, Yiqiang Chen, Yong Liu, Shujun Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add dual-granularity reasoning latent to T2M-GPT | T2M-GPT · HumanML3D | FID: 0.141 -> 0.040 | Improved |
| add dual-granularity reasoning latent to MotionStreamer | MotionStreamer · HumanML3D (continuous) | FID: 11.790 -> 9.937 | Improved |
| add reasoning tokens at 1/4 length versus 1/2 (ablation) | LMR · HumanML3D | gFID: 0.212 -> 0.040 | Improved |
| add reasoning tokens at 1/4 versus 1/8 length (ablation) | LMR · HumanML3D | gFID: 0.062 -> 0.040 | Improved |
| use dual sequences instead of single joint sequence (ablation) | LMR · HumanML3D | gFID: 0.354 -> 0.040 | Improved |
| downsample execution tokens to 1/4 (ablation) | T2M-GPT · HumanML3D | MPJPE: 16.608 -> 30.294 | Hurt |
| two-stage freeze versus end-to-end joint tokenizer training (ablation) | LMR tokenizer · HumanML3D | gFID: 0.145 -> 0.040 | Improved |
| two-stage freeze versus finetuning execution branch (ablation) | LMR tokenizer · HumanML3D | gFID: 0.109 -> 0.040 | Improved |
| shared encoder versus independent networks (ablation) | LMR tokenizer · HumanML3D | gFID: 0.097 -> 0.040 | Improved |
| learned reasoning tokens versus TMR embeddings (ablation) | LMR · HumanML3D | R-Precision Top-1: 0.472 -> 0.537 | Improved |
| reasoning tokens versus coarse-to-fine AR (ablation) | LMR · HumanML3D | FID: 0.154 -> 0.040 | Improved |
| reasoning tokens versus explicit language CoT (ablation) | LMR · HumanML3D | FID: 0.188 -> 0.040 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BAMM — BAMM: Bidirectional Autoregressive Motion Model (ECCV 2024) · better on 1 of 1 shared result
- FG-T2M++ — Fg-T2M: Fine-Grained Text-Driven Human Motion Generation via Diffusion Model (ICCV 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Practice Makes Perfect: From Explicit Decomposition to Reinforced Latent Planning in Text-to-Human Motion (ECCV 2026 · similarity 0.93)
- Motion-R1: Enhancing Motion Generation with Decomposed Chain-of-Thought and RL Binding (ICLR 2026 · similarity 0.91)
- HSI-GPT2: A Dual-Granularity Large Motion Reasoning Model with Diffusion Refinement for Human-Scene Interaction (CVPR 2026 · similarity 0.88)
- AvatarGPT: All-in-One Framework for Motion Understanding Planning Generation and Beyond (CVPR 2024 · similarity 0.88)
- R2C: Mapping Room to Chessboard to Unlock LLM As Low-Level Action Planner (CVPR 2025 · similarity 0.88)
- UniMo: Unified Motion Generation and Understanding with Chain of Thought (AAAI 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
