Research map / 3D vision / Human pose estimation
WAPR: A Foundation Model for Wide-Angle Refinement in Unseen Object Pose Estimation
Yulin Wang, Jianghao Zhou, Hongli Li, MENGTING HU, Chen LUO
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 |
|---|---|---|---|
| train with wide perturbation range ±90° instead of ±20° (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 68.7 -> 81.8 | Improved |
| widen training range to ±180° from ±90° (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 81.8 -> 75.2 | Hurt |
| use SAM-generated masks instead of GT masks in training (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 76.8 -> 81.8 | Improved |
| add mask input (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 80.7 -> 81.8 | Improved |
| add rotational symmetry priors (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 80.1 -> 81.8 | Improved |
| add angle-balanced loss (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 78.2 -> 81.8 | Improved |
| train on SA6D instead of original GSO PBR data (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 76.1 -> 81.8 | Improved |
| attention pooling and transformer encoder instead of average pooling (ablation) | WAPR · IC-BIN/TUD-L/LM-O detection | mean AP: 80.3 -> 81.8 | Improved |
| increase candidates from 4 to 12 (ablation) | WAPR · IC-BIN detection | mean AP: 56.9 -> 70.1 | Improved |
| increase candidates from 12 to 60 (ablation) | WAPR · IC-BIN detection | mean AP: 70.1 -> 72.1 | No clear effect |
| use WAPR as plug-in refiner on FreeZeV2 poses (ablation) | FreeZeV2 · IC-BIN/TUD-L/LM-O detection | mean AP: 83.7 -> 84.3 | Improved |
| use WAPR as plug-in refiner on GigaPose+GenFlow poses (ablation) | GigaPose+GenFlow · IC-BIN/TUD-L/LM-O detection | mean AP: 56.5 -> 57.8 | 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.
- FreeZeV2 — FreeZe: Training-free zero-shot 6D pose estimation with geometric and vision foundation models (ECCV 2024) · better on 2 of 2 shared results
- FoundationPose — FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024) · 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
- Any6D: Model-free 6D Pose Estimation of Novel Objects (CVPR 2025 · similarity 0.85)
- SingRef6D: Monocular Novel Object Pose Estimation with a Single RGB Reference (NeurIPS 2025 · similarity 0.84)
- FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024 · similarity 0.84)
- RefPose: Leveraging Reference Geometric Correspondences for Accurate 6D Pose Estimation of Unseen Objects (CVPR 2025 · similarity 0.84)
- UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image (CVPR 2025 · similarity 0.84)
- Relative Pose Estimation through Affine Corrections of Monocular Depth Priors (CVPR 2025 · similarity 0.83)
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.
