Research map / 3D vision / Human pose estimation
CosyPose: Consistent multi-view multi-object 6D pose estimation
Yann Labbé, Justin Carpentier, Mathieu Aubry, Josef Sivic
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 color data augmentation (ablation) | CosyPose single-view · T-LESS SiSo | recall e_vsd<0.3: 37.0 -> 63.8 | Improved |
| add symmetric disentangled loss instead of point-matching loss (ablation) | CosyPose single-view · T-LESS SiSo | recall e_vsd<0.3: 60.1 -> 63.8 | Improved |
| use EfficientNet base network instead of FlowNet (ablation) | CosyPose single-view · T-LESS SiSo | recall e_vsd<0.3: 59.5 -> 63.8 | Improved |
| use 6D rotation parametrization instead of quaternions (ablation) | CosyPose single-view · T-LESS SiSo | recall e_vsd<0.3: 61.0 -> 63.8 | Improved |
| pretrain on synthetic data (ablation) | CosyPose refinement · YCB-Video single-view | AUC of ADD(-S): approximately 2 points | Improved |
| use CosyPose single-view estimator instead of DeepIM | single-view pose · YCB-Video | AUC of ADD(-S): 81.9 -> 84.5 | Improved |
| use CosyPose instead of Pix2Pose on T-LESS | single-view pose · T-LESS SiSo | recall e_vsd<0.3: 29.5 -> 63.8 | Improved |
| add 5 views instead of 1 view | CosyPose multi-view · YCB-Video | AUC of ADD-S: 89.8 -> 93.4 | Improved |
| add 8 views instead of 1 view | CosyPose multi-view · T-LESS ViVo | mAP@ADD-S<0.1d: 55.0 -> 69.0 | Improved |
| add 4 views instead of 1 view | CosyPose multi-view · T-LESS ViVo | e_vsd<0.3 recall: 62.6 -> 67.6 | Improved |
| add global object-level scene refinement (ablation) | CosyPose multi-view · YCB-Video | ADD-S error (mm): 6.40 -> 5.05 | Improved |
| add global object-level scene refinement (ablation) | CosyPose multi-view · T-LESS | ADD-S error (mm): 4.43 -> 3.19 | 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.
- PVNet — PVNet: Pixel-Wise Voting Network for 6DoF Pose Estimation (CVPR 2019) · better on 1 of 1 shared result
- Pix2Pose — Pix2Pose: Pixel-Wise Coordinate Regression of Objects for 6D Pose Estimation (ICCV 2019) · 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 0.90)
- RefPose: Leveraging Reference Geometric Correspondences for Accurate 6D Pose Estimation of Unseen Objects (CVPR 2025 · similarity 0.89)
- Open-Vocabulary Object 6D Pose Estimation (CVPR 2024 · similarity 0.89)
- One2Any: One-Reference 6D Pose Estimation for Any Object (CVPR 2025 · similarity 0.89)
- Co-op: Correspondence-based Novel Object Pose Estimation (CVPR 2025 · similarity 0.88)
- Pos3R: 6D Pose Estimation for Unseen Objects Made Easy (CVPR 2025 · similarity 0.88)
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.
