Research map / Video / Video generation
Kinematics-Agnostic 3D Human Motion Prediction via Equivariant Latent Diffusion
Cecilia Curreli, Florian Hofherr, Dominik Muhle, Abhishek Saroha, Riccardo Marin, Daniel Cremers
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 |
|---|---|---|---|
| break permutation equivariance by adding positional encoding over joints (ablation) | EquiFusion · H36M zero-shot kinematics | ADE: 0.380 -> 0.467 | Hurt |
| use equivariant architecture instead of joint-dependent weights with multi-dataset training (ablation) | EquiFusion · H36M zero-shot kinematics | ADE: 0.670 -> 0.380 | Improved |
| break equivariance, in-domain AMASS (ablation) | EquiFusion · AMASS | ADE: 0.512 -> 0.539 ADE; MAE 6.525 -> 6.974 | Mixed |
| add explicit joint semantics (limb side or type) (ablation) | EquiFusion · H36M zero-shot kinematics | ADE: 0.380 -> 0.381 | No clear effect |
| add Nymeria to AMASS training data (ablation) | EquiFusion · H36M zero-shot kinematics | uADE: 7.86 -> 7.71 | Improved |
| represent motion as bone directions instead of 3D positions (ablation) | EquiFusion · AMASS | ADE: 0.501 -> 0.498 ADE; CMD 13.963 -> 12.530 | No clear effect |
| represent motion as bone directions instead of 3D positions (ablation) | SkelDiff · AMASS | ADE: 0.480 -> 0.496 ADE; CMD 11.417 -> 9.143 | Mixed |
| natively handle partial kinematics versus heuristic completion | EquiFusion · AMASS with random limb occluded | ADE: 0.567 -> 0.553 (A); 0.499 (A+N) | Improved |
| EquiFusion versus SkelDiff on out-of-distribution MoYoga | EquiFusion · MoYoga | ADE: 0.567 -> 0.501 | Improved |
| EquiFusion versus retargeted baselines in zero-shot H36M kinematics | EquiFusion · H36M zero-shot kinematics | FID: 5.252 -> 0.691 (vs SkelDiff) | Improved |
Closest papers by meaning
- Human Joint Kinematics Diffusion-Refinement for Stochastic Motion Prediction (AAAI 2023 · similarity 0.88)
- Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction (CVPR 2025 · similarity 0.88)
- Pose-Transformed Equivariant Network for 3D Point Trajectory Prediction (CVPR 2024 · similarity 0.85)
- Motion Diversification Networks (CVPR 2024 · similarity 0.85)
- Lifting Motion to the 3D World via 2D Diffusion (CVPR 2025 · similarity 0.84)
- Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration (AAAI 2024 · similarity 0.84)
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.
