Research map / 3D vision / Human pose estimation
LaxMotion: Rethinking Supervision Granularity for 3D Human Motion Generation
Sheng Liu, Yuanzhi Liang, Sidan DU
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 |
|---|---|---|---|
| train with relaxed 2D supervision instead of 3D coordinates | MoMask · HumanML3D | QM Score: 5.850 -> 8.805 | Improved |
| train with relaxed 2D supervision instead of 3D coordinates | MoMask · HumanML3D | FID: 0.045 -> 0.054 | Hurt |
| fuse relaxed-supervision features and finetune with 3D supervision | MoMask · HumanML3D | FID: 0.045 -> 0.034 | Improved |
| train MDM with relaxed supervision | MDM · HumanML3D | FID: 0.544 -> 0.475 | Improved |
| train MoMask with relaxed supervision on KIT-ML | MoMask · KIT-ML | QM Score: 2.504 -> 4.982 | Improved |
| add trajectory-preserving relaxed supervision vs 2D-supervised method | Motion-2-to-3 · HumanML3D global motion | FID: 0.321 -> 0.054 | Improved |
| use relaxed rather than full supervision signal (relaxed input) (ablation) | 3D finetune LaxMotion · HumanML3D | QM Score: 5.879 -> 13.715 | Improved |
| use relaxed rather than full input (ablation) | 3D finetune LaxMotion · HumanML3D | QM Score: 11.200 -> 13.715 | Improved |
| add cross-view plausibility loss L_rec (ablation) | LaxMotion (MoMask-Based) · HumanML3D | FID: 2.588 -> 0.054 | Improved |
| add orientation regularization L_ori (ablation) | LaxMotion (MoMask-Based) · HumanML3D | FID: 0.124 -> 0.054 | Improved |
| add feature consistency regularization L_feat (ablation) | LaxMotion (MoMask-Based) · HumanML3D | FID: 0.088 -> 0.054 | Improved |
| add feature consistency regularization L_feat (ablation) | LaxMotion (MoMask-Based) · HumanML3D | MModality: 2.086 -> 2.046 | Hurt |
Closest papers by meaning
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · similarity 0.88)
- 3D-Aware Implicit Motion Control for View-Adaptive Human Video Generation (CVPR 2026 · similarity 0.88)
- SWSL: Semantic-aware Weakly Supervised Learning for 3D Motion Generation using 2D Motion Data (ECCV 2026 · similarity 0.88)
- Lifting Motion to the 3D World via 2D Diffusion (CVPR 2025 · similarity 0.86)
- Motion Diversification Networks (CVPR 2024 · similarity 0.85)
- 3DTrajMaster: Mastering 3D Trajectory for Multi-Entity Motion in Video Generation (ICLR 2025 · similarity 0.85)
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.
