Research map / 3D vision / Human pose estimation
FoundPose: Unseen Object Pose Estimation with Foundation Features
Evin Pınar Örnek, Yann Labbé, Bugra Tekin, Lingni Ma, Cem Keskin, Christian Forster, Tomas Hodan
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 |
|---|---|---|---|
| use layer 18 instead of the last layer 23 of DINOv2 ViT-L for patch descriptors (ablation) | FoundPose · BOP 7 datasets average | Average AR: 23.5 -> 37.2 | Improved |
| use DINOv2 patch descriptors instead of SAM, DenseSIFT or S2DNet (ablation) | FoundPose · BOP 7 datasets average | Average AR: 37.2 vs 10.7 (SAM), 7.6 (DenseSIFT), 1.1 (S2DNet) | Improved |
| use bag-of-words template retrieval from patch descriptors instead of cls token matching (ablation) | FoundPose · BOP 7 datasets average | Average AR: 18.2 -> 37.2 | Improved |
| mask background black for cls token retrieval (ablation) | cls token retrieval · BOP 7 datasets average | Average AR: 18.2 -> 25.1 | Improved |
| add featuremetric refinement to coarse pose | FoundPose · BOP 7 datasets average | Average AR: 37.2 -> 42.6 | Improved |
| add MegaPose refiner to FoundPose coarse poses | FoundPose · BOP 7 datasets average | Average AR: 37.2 -> 54.7 | Improved |
| refine 5 hypotheses with featuremetric plus MegaPose instead of 1 | FoundPose · BOP 7 datasets average | Average AR: 55.0 -> 59.6 | Improved |
| add featuremetric refinement before MegaPose refinement | FoundPose · BOP 7 datasets average | Average AR: 54.7 -> 55.0 | Improved |
| use ground-truth masks instead of CNOS masks (ablation) | FoundPose · BOP datasets with public GT | AR: +6--19 AR | Improved |
| vary number of templates, PCA components, visual words, RANSAC iterations (ablation) | FoundPose · BOP | Average AR | No clear effect |
| replace ViT-L backbone with ViT-S (layer 9) (ablation) | FoundPose · BOP 7 datasets average | Average AR: 37.2 -> 34.0 | Hurt |
Closest papers by meaning
- UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image (CVPR 2025 · similarity 0.91)
- Co-op: Correspondence-based Novel Object Pose Estimation (CVPR 2025 · similarity 0.89)
- MatchU: Matching Unseen Objects for 6D Pose Estimation from RGB-D Images (CVPR 2024 · similarity 0.89)
- RefPose: Leveraging Reference Geometric Correspondences for Accurate 6D Pose Estimation of Unseen Objects (CVPR 2025 · similarity 0.89)
- Pos3R: 6D Pose Estimation for Unseen Objects Made Easy (CVPR 2025 · similarity 0.89)
- SingRef6D: Monocular Novel Object Pose Estimation with a Single RGB Reference (NeurIPS 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.
