Research map / 3D vision / Human pose estimation
Multi-body SE(3) Equivariance for Unsupervised Rigid Segmentation and Motion Estimation
Read the paperCheck 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 |
|---|---|---|---|
| add global SE(3)-equivariant features to segmentation head (ablation) | segmentation head · SAPIEN | AP: 45.2 -> 51.7 | Helped |
| add point-level invariance flexibility to segmentation head (ablation) | segmentation head · SAPIEN | AP: 51.7 -> 55.3 | Helped |
| update scene flow using only current estimate instead of past flow (ablation) | scene flow updating · SAPIEN | AP: 55.3 -> 54.8 | No clear effect |
| combine past and current flow with a decay factor (ablation) | scene flow updating · SAPIEN | AP: 54.8 -> 57.0 | Helped |
| add motion head consensus score to remove scene flow noise (ablation) | segmentation · SAPIEN | AP: 57.0 -> 63.8 | Helped |
| add motion head consensus score (ablation) | segmentation · SAPIEN | Precision: 63.8 -> 84.2 | Helped |
| add motion head consensus score (ablation) | segmentation · SAPIEN | PQ: 51.6 -> 61.3 | Helped |
| remove part-level equivariance from motion head on unseen categories (ablation) | motion head · SAPIEN val (GT masks) | mean angular error: 5.69 -> 20.77 | Hurt |
| keep part-level equivariance in motion head (ablation) | motion head · SAPIEN val (GT masks) | mean angular error: increases by less than 2 | Helped |
| use SE(3)-equivariant segmentation head vs non-equivariant KPConv (ablation) | segmentation head · SAPIEN pilot | generalization to unseen poses | Helped |
| train with ground-truth labels (Ours-sup) instead of unsupervised | Ours · SAPIEN | AP: 63.8 -> 73.5 | Helped |
| supervised vs unsupervised training | Ours · SAPIEN | EPE3D: 5.47 -> 3.86 | Helped |
Closest papers by meaning
- Equivariant Point Network for 3D Point Cloud Analysis (CVPR 2021 · similarity 0.88)
- Learning Coordinate-based Convolutional Kernels for Continuous SE(3) Equivariant and Efficient Point Cloud Analysis (CVPR 2026 · similarity 0.88)
- Self-Supervised Category-Level Articulated Object Pose Estimation with Part-Level SE(3) Equivariance (ICLR 2023 · similarity 0.88)
- SE(3)-bi-equivariant Transformers for Point Cloud Assembly (NeurIPS 2024 · similarity 0.85)
- World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories (NeurIPS 2026 · similarity 0.85)
- RESfM: Robust Deep Equivariant Structure from Motion (ICLR 2025 · 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.