Research map / 3D vision / Human pose estimation
Generalizing Neural Human Fitting to Unseen Poses With Articulated SE(3) Equivariance
Haiwen Feng, Peter Kulits, Shichen Liu, Michael J. Black, Victoria Fernandez Abrevaya
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 part-based SE(3) equivariant features instead of augmentation-only baselines | SMPL estimation from point clouds · AMASS PosePrior (OOD) | V2V error (cm): PTF aug 6.42 -> ArtEq 3.62 | Improved |
| add global SO(3) data augmentation | PTF · AMASS PosePrior (OOD) part segmentation | accuracy (%): 8.5 -> 80.3 | Improved |
| add global SO(3) data augmentation (ablation) | ArtEq · AMASS PosePrior (OOD) | V2V error (cm): 4.73 -> 3.62 | Improved |
| add parent feature conditioning (ablation) | ArtEq · AMASS PosePrior (OOD) | V2V error (cm): 5.41 -> 4.73 (no aug); 4.17 -> 3.62 (aug) | Improved |
| add parent feature conditioning (ablation) | ArtEq · DFAUST (ID) | V2V error (cm): 0.95 -> 0.98 (aug) | Hurt |
| reduce input points from 5000 to 1000 (ablation) | ArtEq · AMASS PosePrior (OOD) | V2V error (cm): 3.62 -> 4.85 | Hurt |
| remove the pose prior from baselines (ablation) | PTF · AMASS PosePrior (OOD) | V2V error (cm): 6.42 -> 6.46 | No clear effect |
| use regression-only ArtEq instead of test-time optimization | SMPL fitting · inference cost | time (s): 158.1 -> 0.1 | Improved |
| test on raw noisy DFAUST scans without finetuning | ArtEq · DFAUST raw scans (ID poses) | part segmentation accuracy: 96.2 -> 88.3 | Hurt |
Closest papers by meaning
- Self-Supervised Category-Level Articulated Object Pose Estimation with Part-Level SE(3) Equivariance (ICLR 2023 · similarity 0.88)
- Learning to Reconstruct 3D Human Pose and Shape via Model-Fitting in the Loop (ICCV 2019 · similarity 0.88)
- Leveraging SE(3) Equivariance for Self-supervised Category-Level Object Pose Estimation from Point Clouds (NeurIPS 2021 · similarity 0.88)
- Skeleton-Aware 3D Human Shape Reconstruction From Point Clouds (ICCV 2019 · similarity 0.88)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.88)
- Approximately Piecewise E(3) Equivariant Point Networks (ICLR 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.
