Research map / Video / Video generation
Morph: A Motion-free Physics Optimization Framework for Human Motion Generation
Zhuo Li, Mingshuang Luo, Ruibing Hou, Xin Zhao, Hao Liu, Hong Chang, Zimo Liu, Chen Li
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 |
|---|---|---|---|
| train the physics refiner (MPR) on generator-produced motions, with no real motion data (ablation) | MoMask · HumanML3D | physical metrics: PFC 1.058 -> 0.669; Penetrate 23.152 -> 0.0; Float 10.66 -> 2.268; Skate 5.262 -> 0.011 | Improved |
| train MPR on real motion data instead of generated data (ablation) | Morph-MoMask · HumanML3D | FID: 0.074 -> 0.179 | Hurt |
| add VAE prior reward (ablation) | MPR training · HumanML3D | FID: 0.276 -> 0.074 | Improved |
| use the VAE prior reward instead of an adversarial reward (ablation) | MPR training · HumanML3D | FID: 0.194 -> 0.172 | Improved |
| add energy reward (ablation) | MPR training · HumanML3D | FID: 0.163 -> 0.074 | Improved |
| add imitation selection operation (ablation) | Morph-MoMask · HumanML3D | FID: 0.172 -> 0.074 | Improved |
| fine-tune motion generator on physics-refined motions (ablation) | Morph-MoMask · HumanML3D | FID: 0.074 -> 0.041; RTOP-3 0.802 -> 0.816 | Improved |
| add more rounds of alternating MPR and generator training (ablation) | Morph-MoMask · HumanML3D | FID: 0.041 -> 0.034 (three rounds) | Improved |
| raise imitation selection threshold tau from 0.0 to 1.0 (ablation) | Morph-MoMask (no fine-tuning) · HumanML3D | physical vs generation metrics: Skate 5.262 -> 0.003 but FID 0.045 -> 0.084 | Mixed |
| increase amount of generated data for MPR training from 25% to 100% (ablation) | Morph-MoMask (no fine-tuning) · HumanML3D | PFC: 0.866 -> 0.669 | Improved |
| apply Morph to different generators (MDM, MotionDiffuse, T2M-GPT, MoMask) | text-to-motion generators · HumanML3D | Float/Skate: MDM Float 17.502 -> 2.258, Skate 3.540 -> 0.016 | Improved |
| apply Morph to EDGE music-to-dance | EDGE · AIST++ | FID_k: 42.16 -> 39.24 (FID_k); Morph-EDGE dagger 43.01 | Mixed |
Closest papers by meaning
- PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation (NeurIPS 2026 · similarity 0.87)
- CoDA: Coordinated Diffusion Noise Optimization for Whole-Body Manipulation of Articulated Objects (NeurIPS 2025 · similarity 0.87)
- Aligning Human Motion Generation with Human Perceptions (ICLR 2025 · similarity 0.87)
- Optimal-state Dynamics Estimation for Physics-based Human Motion Capture from Videos (NeurIPS 2024 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.86)
- Physical Plausibility-aware Trajectory Prediction via Locomotion Embodiment (CVPR 2025 · similarity 0.86)
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.
