Research map / Video / Video generation
Multi-level Causal LLM-based Text-to-Motion Generation with Human Alignment
Xiaodong Chen, Qian Bao, Xudong Liu, Jianping Fang, Jintao Fang, Yongdong Zhang, Tao Mei, Wu Liu
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 |
|---|---|---|---|
| replace VQ-VAE with Causal RVQ-VAE (ablation) | MoTiGA (step-by-step) · HumanML3D | FID: 0.213 -> 0.186 | Improved |
| use time-synchronized parallel prediction instead of step-by-step (ablation) | MoTiGA · HumanML3D | FID: 0.186 -> 0.064 | Improved |
| use time-lagged causal prediction instead of time-synchronized (ablation) | MoTiGA · HumanML3D | FID: 0.064 -> 0.055 | Improved |
| add GRPO preference alignment (ablation) | MoTiGA · HumanML3D | FID: 0.055 -> 0.047 | Improved |
| use MHPO multi-level reward instead of GRPO (ablation) | MoTiGA · HumanML3D | FID: 0.047 -> 0.041 | Improved |
| use Causal RVQ-VAE instead of RVQ-VAE at 4 levels (ablation) | MoTiGA · HumanML3D | FID (generation): 0.085 -> 0.055 | Improved |
| use Causal RVQ-VAE instead of RVQ-VAE (ablation) | tokenizer · HumanML3D | FID (reconstruction): 0.027 -> 0.031 at V=4 | Hurt |
| increase quantization levels from 4 to 7 (ablation) | Causal RVQ-VAE · HumanML3D | FID (generation): 0.055 -> 0.064 | No clear effect |
| set MHPO bonus weight to 0.6 instead of 0.0 or 1.0 (ablation) | MHPO · HumanML3D | FID: 0.047 (0.0) -> 0.041 (0.6); 0.044 (1.0) | Improved |
| add given initial pose condition | MoTiGA · HumanML3D | FID: 0.041 -> 0.040 | No clear effect |
Closest papers by meaning
- MOGO: Residual Quantized Hierarchical Causal Transformer for Real-Time and Infinite-Length 3D Human Motion Generation (AAAI 2026 · similarity 0.91)
- MoLingo: Motion-Language Alignment for Text-to-Human Motion Generation (CVPR 2026 · similarity 0.90)
- LLaMo: Scaling Pretrained Language Models for Unified Motion Understanding and Generation with Continuous Autoregressive Tokens (CVPR 2026 · similarity 0.90)
- GenM3: Generative Pretrained Multi-path Motion Model for Text Conditional Human Motion Generation (ICCV 2025 · similarity 0.89)
- MotionRFT: Unified Reinforcement Fine-Tuning for Text-to-Motion Generation (TPAMI 2026 · similarity 0.89)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · 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.
