Research map / 3D vision / 3D generation
SWSL: Semantic-aware Weakly Supervised Learning for 3D Motion Generation using 2D Motion Data
Zhicheng Shi, Tuo Feng, Wenguan Wang, Yi Yang
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 SWSL training | SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.577 | Improved |
| add SWSL training | Momask* · HumanML3D (TMR protocol) | FID: 0.595 -> 0.539 | Improved |
| add SWSL training | DiP* · HumanML3D (TMR protocol) | FID: 0.736 -> 0.698 | Improved |
| add SWSL training | SB · HumanML3D (CONVENTION protocol) | FID: 0.109 -> 0.097 | Improved |
| add SWSL training | Momask* · HumanML3D (CONVENTION protocol) | FID: 0.089 -> 0.079 | Improved |
| add SWSL training | DiP* · HumanML3D (CONVENTION protocol) | FID: 0.283 -> 0.266 | Improved |
| add SPE pseudo-label enhancement alone (ablation) | SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.586 | Improved |
| add SFO feedback optimization alone (ablation) | SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.590 | Improved |
| combine SPE and SFO (full SWSL) (ablation) | SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.577 | Improved |
| increase 2D motion data from 3k to 12k (ablation) | SWSL on SB · HumanML3D (TMR protocol) | FID: 0.613 -> 0.574 | Improved |
| reduce 3D data to 1/4 while adding 2D data (ablation) | SWSL on SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.632 | No clear effect |
| reduce 3D data to 1/8 while adding 2D data (ablation) | SWSL on SB · HumanML3D (TMR protocol) | FID: 0.625 -> 0.691 | Hurt |
Closest papers by meaning
- Motion-2-to-3: Leveraging 2D Motion Data for 3D Motion Generations (ICCV 2025 · similarity 0.89)
- Hierarchical Enhancement of Semantic Priors for Disentangled Text-Driven Motion Generation (CVPR 2026 · similarity 0.88)
- Generating Diverse and Natural 3D Human Motions From Text (CVPR 2022 · similarity 0.88)
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · similarity 0.87)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.87)
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.87)
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.
