Research map / 3D vision / Human pose estimation
Pose without Guesses: Generalizable Object Pose Estimation From a Single Reference
Chen Zhao, Tong Zhang, Zheng Dang, Mathieu Salzmann
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 WCV rotation estimation instead of a rotation regression module (ablation) | DVMNet++ · CO3D | mean angular error: 31.78 -> 19.95 | Improved |
| add voxel weights (2D mask and objectness) to closest voxel algorithm (ablation) | DVMNet++ · CO3D | mean angular error: 21.64 -> 19.95 | Improved |
| add 2D object mask weights to WCV (ablation) | DVMNet++ · CO3D | mean angular error: 21.64 -> 20.92 | Improved |
| add voxel objectness weights to WCV (ablation) | DVMNet++ · CO3D | mean angular error: 21.64 -> 20.07 | Improved |
| add voxel objectness on top of 2D mask (ablation) | DVMNet++ · CO3D | mean angular error: 20.92 -> 19.95 | Improved |
| add image-retrieval (DINOv2 similarity) proposal selection to detector (ablation) | open-set detector · LINEMOD | translation angular error: 5.27 -> 1.55 | Improved |
| replace GT bounding box with predicted box from open-set detector | DVMNet++ · CO3D | mean angular error: 19.95 -> 19.64 | No clear effect |
| replace GT bounding box with predicted box from open-set detector | DVMNet++ · LINEMOD | mean angular error: 36.82 -> 39.79 | Hurt |
| increase number of reference images from 1 to 7 (ablation) | DVMNet++ · CO3D | mean angular error: 19.95 -> 14.93 | Improved |
| increase number of reference images from 1 to 7 (ablation) | 3DAHV · CO3D | mean angular error: 28.44 -> 24.76 | Improved |
| increase number of reference images from 1 to 7 (ablation) | SuperGlue · CO3D | mean angular error: 71.47 -> 61.74 | Improved |
| use hypothesis-free DVMNet++ instead of hypothesis-based 3DAHV | pose estimator · CO3D | time per image pair: 35ms -> 23ms | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 5 of 5 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
- DVMNet: Computing Relative Pose for Unseen Objects Beyond Hypotheses (CVPR 2024 · similarity 0.93)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (ICLR 2024 · similarity 0.93)
- Deep Learning-Based Object Pose Estimation: A Comprehensive Survey (IJCV 2026 · similarity 0.92)
- UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image (CVPR 2025 · similarity 0.92)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.90)
- Co-op: Correspondence-based Novel Object Pose Estimation (CVPR 2025 · similarity 0.89)
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.
