Research map / 3D vision / 3D generation
Towards Natural and Accurate Future Motion Prediction of Humans and Animals
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 Lie algebra skeleton representation instead of independent joint coordinates | motion prediction · H3.6m long-term walking | naturalness of predicted motion | Helped |
| use hierarchical local and sequence-level motion states (ablation) | HMR · H3.6m, predictions beyond 400ms | MAE: 1.89 vs 1.72; 1.95 vs 1.73; 2.65 vs 2.50; 2.06 vs 1.84 | Helped |
| add left and right forget gates for neighboring frame states (ablation) | HMR · H3.6m at 1000ms | MAE: 1.80 -> 1.72; 1.85 -> 1.73; 2.60 -> 2.50; 2.00 -> 1.84 | Helped |
| add a second decoder layer (ablation) | HMR · H3.6m at 1000ms | MAE: 1.86 -> 1.72; 1.88 -> 1.73; 2.66 -> 2.50; 2.07 -> 1.84 | Helped |
| use kinematic-chain-weighted loss instead of L2 loss (ablation) | HMR · H3.6m | MJE: 113.4 -> 46.4 | Helped |
| use kinematic-chain-weighted loss instead of L2 loss (ablation) | HMR · H3.6m | MAE: 1.40 -> 1.36 | Helped |
| increase hidden state size from 100 to 300 (ablation) | HMR · H3.6m validation | validation loss: 0.177 -> 0.151 | Helped |
| increase hidden state size from 300 to 400 or 500 (ablation) | HMR · H3.6m validation | validation loss: 0.151 -> 0.172; 0.151 -> 0.175 | Hurt |
| increase recurrent steps from 1 to 10 (ablation) | HMR · H3.6m validation | validation loss: 0.169 -> 0.151 | Helped |
| increase recurrent steps from 10 to 15 or 20 (ablation) | HMR · H3.6m validation | validation loss: 0.151 -> 0.159; 0.151 -> 0.161 | Hurt |
| enlarge context window from 3 to 5, 7, or 11 (ablation) | HMR · H3.6m validation | validation loss: 0.151 -> 0.155; 0.151 -> 0.159; 0.151 -> 0.158 | Hurt |
| use HMR rather than the strongest listed baseline | motion prediction · H3.6m Discussion, Greeting, Posing, and Walking Dog at 1000ms | MAE: 1.88 -> 1.72; 1.80 -> 1.73; 2.51 -> 2.50; 1.90 -> 1.84 | Helped |
Closest papers by meaning
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.88)
- ALIEN: Implicit Neural Representations for Human Motion Prediction under Arbitrary Latency (CVPR 2025 · similarity 0.87)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.87)
- Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration (AAAI 2024 · similarity 0.86)
- Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction (CVPR 2025 · similarity 0.86)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.86)
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.