Research map / 3D vision / Human pose estimation
Geometric Neural Distance Fields for Learning Human Motion Priors
Zhengdi Yu, Simone Foti, Linguang Zhang, Amy Zhao, Cem Keskin, Stefanos Zafeiriou, Tolga Birdal
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 transition and acceleration priors on top of pose prior (NRMF vs NRDF-t) (ablation) | NRMF · AMASS motion denoising | Pos. Err. All (mm): 22.8 -> 16.4 | Improved |
| add transition prior T-NRDF to SMPLer-X refinement (ablation) | SMPLer-X + NRDF · 3DPW 2D fitting | Acc Err (m/s^2): 24.31 -> 9.89 | Improved |
| add acceleration prior A-NRDF to SMPLer-X refinement (ablation) | SMPLer-X + NRDF · 3DPW 2D fitting | Acc Err (m/s^2): 24.31 -> 6.73 | Improved |
| combine pose, transition and acceleration priors (NRMF full) versus pose prior only (ablation) | SMPLer-X refinement · 3DPW 2D fitting | MPJPE (mm): 71.88 -> 66.13 | Improved |
| optimize SMPLer-X with no prior (ablation) | SMPLer-X · 3DPW 2D fitting | MPJPE (mm): 82.65 -> 84.67 | Hurt |
| use Riemannian-geometry NRDF-based transition prior instead of Euclidean PoseNDF (ablation) | NRDF/PoseNDF pose prior · AMASS motion denoising | Pos. Err. All (mm): 18.8 -> 16.7 | Improved |
| conditioned product-space NRMF instead of decoupled Motion-NDF (ablation) | NRMF · AMASS motion denoising | Acc Err: 2.64 -> 2.25 | Improved |
| use pose prior only (-t) for partial observations (ablation) | NRDF-t · AMASS partial 3D observations | Pos. Err. Vis vs Occ (mm): Vis 4.2 but Occ 194.5 | Mixed |
| NRMF versus RoHM on partial observations | NRMF · AMASS partial 3D observations | Pos. Err. All (mm): 51.3 -> 46.3 | Improved |
| NRMF versus RoHM on PROX RGB frequency of floor penetration distance (RGB) | NRMF · PROX RGB | Dist: 14.31 -> 16.21 | Hurt |
| NRMF versus HuMoR on motion generation | NRMF · Motion generation | FID_m: 9.147 -> 5.317 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- RoHM — RoHM: Robust Human Motion Reconstruction via Diffusion (CVPR 2024) · better on 6 of 6 shared results
- HuMoR — HuMoR: 3D Human Motion Model for Robust Pose Estimation (ICCV 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- NRDF: Neural Riemannian Distance Fields for Learning Articulated Pose Priors (CVPR 2024 · similarity 0.91)
- A Neural Temporal Model for Human Motion Prediction (CVPR 2019 · similarity 0.89)
- NIFTY: Neural Object Interaction Fields for Guided Human Motion Synthesis (CVPR 2024 · similarity 0.88)
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · similarity 0.87)
- Degrees of Freedom Matter: Inferring Dynamics from Point Trajectories (CVPR 2024 · similarity 0.87)
- SurMo: Surface-based 4D Motion Modeling for Dynamic Human Rendering (CVPR 2024 · similarity 0.87)
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.
