Research map / Graph learning / Knowledge and temporal graphs
Dynamic Multiscale Graph Neural Networks for 3D Skeleton Based Human Motion Prediction
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use adaptive multiscale graphs and cross-scale feature fusion | DMGNN · Human 3.6M, short-term average | MAE: 0.27, 0.52, 0.83, 0.95 | Helped |
| Use DMGNN rather than Traj-GCN | Human 3.6M, long-term average · Human 3.6M, 560 ms and 1000 ms | MAE: 0.90 -> 0.89; 1.27 -> 1.21 | Mixed |
| Replace adaptive single-scale graphs with fixed graphs (ablation) | DMGNN · Human 3.6M, short-term representative actions | MAE: 0.18 -> 0.20 (Walking, 80 ms); 0.17 -> 0.20 (Eating, 80 ms) | Hurt |
| Remove graph-based GRU (ablation) | DMGNN · Human 3.6M, short-term representative actions | MAE: 0.18 -> 0.22 (Walking, 80 ms); 0.17 -> 0.19 (Eating, 80 ms) | Hurt |
| Add multiple body scales instead of using only joint scale (ablation) | DMGNN · Human 3.6M, short-term average, 400 ms | MAE: 1.00 -> 0.95 | Helped |
| Add very abstract body scales (s4 or s5) (ablation) | DMGNN · Human 3.6M, short-term average, 400 ms | MAE: 0.95 -> 0.98; 0.95 -> 0.99 | Hurt |
| Increase encoder depth from 1 to 4 MGCUs (ablation) | DMGNN · Human 3.6M, short- and long-term | MAE: 1.02 -> 0.95 (400 ms); 1.52 -> 1.48 (1000 ms) | Helped |
| Increase depth beyond 4 MGCUs (ablation) | DMGNN · Human 3.6M, short- and long-term | running time: 29.18 -> 30.37 (400 ms); 86.04 -> 88.39 (1000 ms) | Hurt |
| Add relative features when inferring cross-scale graphs (ablation) | DMGNN CS-FBs · Human 3.6M, 400 ms | average MAE: 0.630 -> 0.613 | Helped |
| Use 2 CS-FBs rather than none (ablation) | DMGNN · Human 3.6M, 400 ms | average MAE: 0.630 -> 0.613 | Helped |
| Use too many CS-FBs (ablation) | DMGNN · Human 3.6M, 400 ms | average MAE: 0.613 -> 0.619 | Hurt |
| Include velocity and acceleration differences along with pose (ablation) | DMGNN encoder and decoder · Human 3.6M, short-term prediction | MAE: 1.01 -> 0.95 (400 ms); 0.34 -> 0.27 (80 ms) | Helped |
Closest papers by meaning
- Learning Dynamic Relationships for 3D Human Motion Prediction (CVPR 2020 · similarity 0.89)
- Multi-Scale Spatial Temporal Graph Convolutional Network for Skeleton-Based Action Recognition (AAAI 2021 · similarity 0.88)
- Skeleton-Based Action Recognition With Directed Graph Neural Networks (CVPR 2019 · similarity 0.87)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.86)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.86)
- BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition (CVPR 2024 · 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.