Research map / 3D vision / Face and shape
Deformation-Aware 3D Model Embedding and Retrieval
Mikaela Angelina Uy, Jingwei Huang, Minhyuk Sung, Tolga Birdal, Leonidas Guibas
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 the egocentric distance field to the margin-loss model (replace identity G) (ablation) | Ours-Margin · ShapeNet Table | e_D^m top-1: 2.183 -> 2.127 | Improved |
| add the egocentric distance field to the regression model (replace identity G) (ablation) | Ours-Reg · ShapeNet Table | e_D^m top-1: 2.500 -> 1.969 | Improved |
| add the egocentric distance field to the regression model (replace identity G) (ablation) | Ours-Reg · ShapeNet Chair | e_D^m top-1: 4.349 -> 1.752 | Improved |
| define shape relations by fitting gap instead of Chamfer distance (regression loss, identity G) (ablation) | Symm-Reg · ShapeNet Table | e_D^m top-1: 5.086 -> 2.500 | Improved |
| define shape relations by fitting gap instead of Chamfer distance (margin loss, identity G) (ablation) | Symm-Margin · ShapeNet Table | e_D^m top-1: 2.362 -> 2.183 | Improved |
| use regression loss instead of margin loss with egocentric distance field | Ours · ShapeNet Table | e_D^m top-1: 2.127 -> 1.969 | Improved |
| use regression loss instead of margin loss without egocentric field (identity G) | Symm-Margin · ShapeNet Chair | e_D^m top-1: 1.946 -> 4.349 | Hurt |
| use deformation-aware Ours-Reg instead of Chamfer ranking for retrieval | Ranked CD · ShapeNet Table | e_D^m top-1: 2.095 -> 1.969 | Improved |
| use Ours-Reg instead of Ranked-CD for retrieval (150-model ranking) | Ranked-CD · ShapeNet Table | mean rank of top-1: 12.53 -> 7.05 | Improved |
| use Ours-Reg instead of Ranked-CD for retrieval | Ranked-CD · ShapeNet Table | recall@1: 50.50 -> 70.64 | Improved |
| use Ours-Reg instead of Ranked CD on planes | Ranked CD · ShapeNet Plane | e_D^m top-1: 1.199 -> 1.199 | No clear effect |
| use Ours-Reg instead of Ranked-CD in the car ranking | Ranked-CD · ShapeNet Car | mean rank of top-1: 12.0 -> 7.80 | Improved |
Closest papers by meaning
- Joint Learning of 3D Shape Retrieval and Deformation (CVPR 2021 · similarity 0.92)
- KP-RED: Exploiting Semantic Keypoints for Joint 3D Shape Retrieval and Deformation (CVPR 2024 · similarity 0.88)
- 3D-CODED: 3D Correspondences by Deep Deformation (ECCV 2018 · similarity 0.87)
- 3D Pose Estimation and 3D Model Retrieval for Objects in the Wild (CVPR 2018 · similarity 0.87)
- Unsupervised Dense Deformation Embedding Network for Template-Free Shape Correspondence (ICCV 2021 · similarity 0.87)
- DeepShape: Deep-Learned Shape Descriptor for 3D Shape Retrieval (TPAMI 2016 · 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.
