Research map / Video / Video generation
Causal Motion Diffusion Models for Autoregressive Motion Generation
Qing Yu, Akihisa Watanabe, Kent Fujiwara
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 MAC-VAE with standard VAE (ablation) | CMDM · HumanML3D | FID: 0.068 -> 0.107 | Hurt |
| Add motion-language alignment to causal VAE (ablation) | CMDM · HumanML3D | R-Top1: 0.575 -> 0.588 | Improved |
| Add motion-language alignment to causal VAE (ablation) | CMDM · HumanML3D | FID: 0.070 -> 0.068 | No clear effect |
| Use causal diffusion forcing instead of full-sequence diffusion (ablation) | CMDM · HumanML3D transition | AUJ: 0.72 -> 0.42 | Improved |
| Use causal diffusion forcing instead of full-sequence diffusion (ablation) | CMDM · HumanML3D | R-Top1: 0.591 -> 0.588 | No clear effect |
| Add ROPE (ablation) | CMDM · HumanML3D | FID: 0.087 -> 0.068 | Improved |
| Add AdaLN and ROPE together (ablation) | CMDM · HumanML3D transition | FID: 2.33 -> 1.66 | Improved |
| Add frame-wise sampling schedule (FSS) (ablation) | CMDM · HumanML3D | FID: 0.079 -> 0.068 | Improved |
| Add frame-wise sampling schedule (FSS) (ablation) | CMDM · inference speed | fps: 28 -> 125 | Improved |
| Use larger uncertainty scale L=10 (K=20) (ablation) | CMDM FSS · HumanML3D | FID: 0.068 -> 0.103 | Hurt |
| Use uncertainty scale L=5 (K=50) (ablation) | CMDM FSS · HumanML3D transition | AUJ: 0.42 -> 0.38 | Improved |
| Use CMDM vs MARDM on long-horizon | CMDM · SnapMoGen long-horizon | FID: 40.80 -> 32.49 | Improved |
Closest papers by meaning
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.91)
- Generative Pre-trained Autoregressive Diffusion Transformer (NeurIPS 2025 · similarity 0.90)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.90)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.90)
- Towards Detailed Text-to-Motion Synthesis via Basic-to-Advanced Hierarchical Diffusion Model (AAAI 2024 · similarity 0.90)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.89)
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.
