Research map / 3D vision / Human pose estimation
Pseudo-keypoint RKHS Learning for Self-supervised 6DoF Pose Estimation
Yangzheng Wu, Michael Alan Greenspan
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 dense multi-layer feature matching in the Adapter instead of sparse matching of the radial map only (ablation) | RKHSPose · LM + five BOP core datasets | AR: 77.9 -> 84.4 | Improved |
| alternate real and synthetic epochs instead of sequential training (ablation) | RKHSPose · LM + five BOP core datasets | AR: 81.9 -> 84.4 | Improved |
| add trainable weights to the linear kernel (ablation) | RKHS Adapter · LM + five BOP core datasets | AR: 71.0 -> 84.4 | Improved |
| use RBF kernel instead of linear kernel (both with trainable weights) (ablation) | RKHS Adapter · LM + five BOP core datasets | AR: 84.4 -> 81.6 | Hurt |
| use MMD instead of KL divergence as domain gap loss (ablation) | RKHS Adapter · LM + five BOP core datasets | AR: 77.9 -> 84.4 | Improved |
| use MMD instead of Wasserstein distance (ablation) | RKHS Adapter · LM + five BOP core datasets | AR: 80.8 -> 84.4 | Improved |
| add unlabeled real images with pseudo-keypoint adapter training to synthetic-only training (ablation) | RKHSPose · LINEMOD | ADD(S): 78.2 -> 95.8 | Improved |
| increase number of real images beyond 160-320 (ablation) | RKHSPose · BOP datasets | AR: +0.1% | No clear effect |
| add ICP refinement (ablation) | RKHSPose · YCB-Video | ADD(S) AUC: 82.8 -> 83.0 | Improved |
| use RKHSPose without real GT labels versus fully supervised methods | BOP Occlusion LINEMOD · Occlusion LINEMOD | ADD(S): 68.2 vs 79.7 (RADet+PFA) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Self6D++ — Self6D: Self-Supervised Monocular 6D Object Pose Estimation (ECCV 2020) · better on 3 of 3 shared results
- TexPose — TexPose: Neural Texture Learning for Self-Supervised 6D Object Pose Estimation (CVPR 2023) · better on 2 of 2 shared results
- SMOC-Net — SMOC-Net: Leveraging Camera Pose for Self-Supervised Monocular Object Pose Estimation (CVPR 2023) · better on 1 of 1 shared result
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
- SO-Pose: Exploiting Self-Occlusion for Direct 6D Pose Estimation (ICCV 2021 · similarity 0.88)
- Pos3R: 6D Pose Estimation for Unseen Objects Made Easy (CVPR 2025 · similarity 0.85)
- One2Any: One-Reference 6D Pose Estimation for Any Object (CVPR 2025 · similarity 0.85)
- FAR: Flexible Accurate and Robust 6DoF Relative Camera Pose Estimation (CVPR 2024 · similarity 0.85)
- GIVEPose: Gradual Intra-class Variation Elimination for RGB-based Category-Level Object Pose Estimation (CVPR 2025 · similarity 0.85)
- Instance-Adaptive and Geometric-Aware Keypoint Learning for Category-Level 6D Object Pose Estimation (CVPR 2024 · similarity 0.84)
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.
