Research map / 3D vision / Human pose estimation
CoMusion: Towards Consistent Stochastic Human Motion Prediction via Motion Diffusion
Jiarui Sun, Girish Chowdhary
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 Transformer-based reconstruction module F in front of GCN refinement (ablation) | CoMusion · Human3.6M | ADE: 0.539 -> 0.382 | Improved |
| add GCN-DCT refinement module R (ablation) | CoMusion · Human3.6M | ADE: 0.502 -> 0.382 | Improved |
| use proposed variance scheduler (offset 1, alpha_bar_0 = 0.5) instead of standard cosine (ablation) | CoMusion · Human3.6M | FID: 0.282 -> 0.102 | Improved |
| predict y0 directly instead of noise epsilon (ablation) | CoMusion · Human3.6M | CMD: 25.968 -> 3.202 | Improved |
| use proposed scheduler instead of sqrt scheduler (ablation) | CoMusion · Human3.6M | FID: 0.128 -> 0.102 | Improved |
| use linear scheduler instead of proposed (ablation) | CoMusion · Human3.6M | training stability | Hurt |
| reconstruct motion history in loss (gamma > 0) instead of gamma = 0 (ablation) | CoMusion · Human3.6M | FID: 0.123 -> 0.102 | Improved |
| use l1 loss instead of l2 loss (ablation) | CoMusion · Human3.6M | CMD: 8.215 -> 3.202; APD 9.054 -> 7.632 | Mixed |
| relaxation k = 2 instead of k = 1 (ablation) | CoMusion · Human3.6M | APD: 4.061 -> 7.632 | Improved |
| relaxation k = 3 instead of k = 2 (ablation) | CoMusion · Human3.6M | FID: 0.102 -> 0.191 | Hurt |
| CoMusion vs prior state of the art (consistency) | CoMusion · Human3.6M | CMD / FID: CMD 4.927 -> 3.202; FID 0.209 -> 0.102 | Improved |
| CoMusion vs diversity-focused baselines | CoMusion · Human3.6M | APD: 7.632 vs 15.353 (MotionDiff) | Hurt |
Closest papers by meaning
- A Stochastic Conditioning Scheme for Diverse Human Motion Prediction (CVPR 2020 · similarity 0.92)
- Generating Smooth Pose Sequences for Diverse Human Motion Prediction (ICCV 2021 · similarity 0.91)
- Harmonizing Stochasticity and Determinism: Scene-responsive Diverse Human Motion Prediction (NeurIPS 2024 · similarity 0.89)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.89)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.88)
- Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction (CVPR 2025 · similarity 0.88)
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.
