Research map / 3D vision / Human pose estimation
ConceptPose: Training-Free Zero-Shot Object Pose Estimation using Concept Vectors
Liming Kuang, Yordanka Velikova, Mahdi Saleh, Jan-Nico Zaech, Danda Pani Paudel, Benjamin Busam
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 |
|---|---|---|---|
| increase number of concepts from 3 to 20 (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 64.3 -> 74.0 | Improved |
| increase number of concepts from 15 to 20 (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 72.0 -> 74.0 | Improved |
| use continuous concept vectors instead of binary masks (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 30.5 -> 71.5 | Improved |
| use SigLIP2-giant instead of SigLIP2-large (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 67.0 -> 72.0 | Improved |
| use SigLIP2-giant instead of CLIP ViT-L/14@336px (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 54.1 -> 72.0 | Improved |
| use SigLIP2-giant GradCAM instead of DINOv3 + dinotxt patch similarity (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 62.3 -> 72.0 | Improved |
| add voxelization (64^3 grid) (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 71.5 -> 71.2 | Mixed |
| add voxelization (64^3 grid) (ablation) | ConceptPose · LINEMOD | ADD(-S) recall: 38.6 -> 36.8 | Mixed |
| use geometric prompt with category renderings instead of default part-name prompt (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 71.5 -> 71.9 | No clear effect |
| use affordance prompt instead of default prompt (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 71.5 -> 68.6 | Hurt |
| use cosine instead of bidirectional KL correspondence (ablation) | ConceptPose · REAL275 | ADD(-S) recall: 72.0 -> 72.1 | No clear effect |
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.
- Any6D — Any6D: Model-free 6D Pose Estimation of Novel Objects (CVPR 2025) · better on 4 of 4 shared results
- One2Any — One2Any: One-Reference 6D Pose Estimation for Any Object (CVPR 2025) · better on 4 of 4 shared results
- FoundationPose — FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024) · better on 2 of 2 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
- Universal Features Guided Zero-Shot Category-Level Object Pose Estimation (AAAI 2025 · similarity 0.89)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.88)
- UNOPose: Unseen Object Pose Estimation with an Unposed RGB-D Reference Image (CVPR 2025 · similarity 0.87)
- Open-Vocabulary Object 6D Pose Estimation (CVPR 2024 · similarity 0.87)
- Vision Foundation Model Enables Generalizable Object Pose Estimation (NeurIPS 2024 · similarity 0.87)
- SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation (CVPR 2024 · similarity 0.87)
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.
