Research map / 3D vision / Human pose estimation
Forecasting Characteristic 3D Poses of Human Actions
Christian Diller, Thomas Funkhouser, Angela Dai
Check 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 |
|---|---|---|---|
| Use probabilistic volumetric heatmap predictions instead of a deterministic MLP head (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.064 -> 0.054 | Improved |
| Use probabilistic volumetric heatmap predictions instead of a deterministic MLP head (ablation) | Characteristic 3D Pose Forecasting · Human3.6M | MPJPE: 0.108 -> 0.092 | Improved |
| Sample joints autoregressively rather than independently (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.077 -> 0.054 | Improved |
| Add per-voxel offset prediction (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.132 -> 0.054 | Improved |
| Add pose refinement (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.057 -> 0.054 | Improved |
| Increase pose samples from the default to k=50 (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.054 -> 0.049 | Improved |
| Use discretized cross-entropy heatmap loss rather than ℓ1 regression loss (ablation) | Characteristic 3D Pose Forecasting · Human3.6M | MPJPE: 0.198 -> 0.092 | Improved |
| Use discretized cross-entropy heatmap loss rather than ℓ2 regression loss (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.130 -> 0.054 | Improved |
| Use right hand → left hand → rest autoregressive ordering (ablation) | Characteristic 3D Pose Forecasting · GRAB | Diversity: 0.049 -> 0.105 | Improved |
| Use a 16³ grid with offset prediction rather than a 32³ grid with offsets (ablation) | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.066 -> 0.054 | Improved |
| Provide ground-truth action labels as additional input (ablation) | Characteristic 3D Pose Forecasting · GRAB | IS: 4.153 { $\pm 0.87$} -> 1.085 { $\pm 0.02$} | Hurt |
| Use the proposed characteristic-pose method rather than DLow | Characteristic 3D Pose Forecasting · GRAB | MPJPE: 0.071 -> 0.054 | Improved |
Closest papers by meaning
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.91)
- Predicting 3D Human Dynamics From Video (ICCV 2019 · similarity 0.89)
- Forecasting of 3D Whole-body Human Poses with Grasping Objects (CVPR 2024 · similarity 0.89)
- Expressive Forecasting of 3D Whole-Body Human Motions (AAAI 2024 · similarity 0.88)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.87)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · 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.
