Research map / 3D vision / 3D generation
Zero-shot Inexact CAD Model Alignment from a Single Image
Pattaramanee Arsomngern, Sasikarn Khwanmuang, Matthias Nießner, Supasorn Suwajanakorn
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 geometry-aware adapter features fused with DINOv2 in coarse alignment (ablation) | coarse alignment · ScanNet25k | avg instance accuracy: 18.8 -> 26.0 | Improved |
| fuse DINOv2 features with adapter features instead of using adapter features only (omega=1) (ablation) | coarse alignment · ScanNet25k | avg instance accuracy: 21.0 -> 26.0 | Improved |
| add NOC dense alignment refinement after coarse alignment (ablation) | Ours · ScanNet25k | avg instance accuracy: 26.0 -> 30.1 | Improved |
| use NOC dense alignment instead of feature-metric (FM) dense alignment (ablation) | Ours · ScanNet25k | avg instance accuracy: 26.1 -> 30.1 | Improved |
| use feature-metric dense alignment on top of DINOv2 coarse alignment (ablation) | DINOv2 coarse · ScanNet25k | avg instance accuracy: 18.8 -> 18.8 | No clear effect |
| use nearest neighbor matching in adapter feature space instead of directly regressed NOC (NOC-S) for correspondences (ablation) | coarse + NOC refinement · ScanNet25k | avg instance accuracy: 19.9 -> 30.1 | Improved |
| replace DINOv2 with our features in coarse step before NOC refinement (ablation) | NOC refinement pipeline · ScanNet25k | avg instance accuracy: 24.2 -> 30.1 | Improved |
| add depth loss to dense alignment (alone) (ablation) | dense alignment · ScanNet25k 9 categories | avg instance accuracy: 23.93 -> 24.87 | Improved |
| add mask loss alone to dense alignment (ablation) | dense alignment · ScanNet25k 9 categories | avg instance accuracy: 23.93 -> 18.92 | Hurt |
| add NOC alignment loss alone to dense alignment (ablation) | dense alignment · ScanNet25k 9 categories | avg instance accuracy: 23.93 -> 23.76 | No clear effect |
| combine NOC-A, mask and depth losses in dense alignment (ablation) | dense alignment · ScanNet25k 9 categories | avg instance accuracy: 23.93 -> 28.36 | Improved |
| adapt our method to 6-DoF with ground-truth scales versus FoundationPose | Ours (for 6D) · ScanNet25k | avg instance accuracy: 31.1 -> 38.0 | 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.
- FoundationPose — FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects (CVPR 2024) · better on 4 of 4 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
- Vid2CAD: CAD Model Alignment Using Multi-View Constraints From Videos (TPAMI 2022 · similarity 0.87)
- Scan2CAD: Learning CAD Model Alignment in RGB-D Scans (CVPR 2019 · similarity 0.87)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.85)
- GeoReF: Geometric Alignment Across Shape Variation for Category-level Object Pose Refinement (CVPR 2024 · similarity 0.85)
- MRC-Net: 6-DoF Pose Estimation with MultiScale Residual Correlation (CVPR 2024 · similarity 0.84)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.84)
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.
