Research map / 3D vision / Face and shape
Geometry-Guided Modeling of Foundation Features Enables Generalizable Object Shape Deformation Learning
Yiyao Ma, Kai Chen, Zhongxiang Zhou, Zhuheng Song, Dongsheng Xie, Zelong Tan, Rong Xiong, Qi Dou
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 flow matching instead of direct regression (ablation) | Our-MV · Novel objects (Retrieved template) | S-IoU: 45.78 -> 48.79 | Improved |
| add flow matching instead of direct regression (ablation) | Our-MV · Novel objects (Random template) | CD: 2.74 -> 2.46 | Improved |
| add feature propagation to all template points (ablation) | Our-MV · Novel objects (Random template) | S-IoU: 41.10 -> 47.31 | Improved |
| add relational modeling between template and target features (ablation) | Our-MV · Novel objects (Random template) | S-IoU: 44.67 -> 47.31 | Improved |
| add primary view selection in aggregation (ablation) | Our-MV · Novel objects (Retrieved template) | S-IoU: 44.50 -> 48.79 | Improved |
| add camera pose encoding in multi-view aggregation (ablation) | Our-MV · Novel objects (Retrieved template) | S-IoU: 44.98 -> 48.79 | Improved |
| naive multi-view fusion without primary view selection or pose encoding vs single view (ablation) | Our-SV · Novel objects (Retrieved template) | S-IoU: 48.45 -> 44.50 | Hurt |
| use multi-view template fusion instead of single view | Our-SV · Seen categories (Random template) | S-IoU: 46.78 -> 47.31 | Improved |
| use multi-view template fusion instead of single view | Our-SV · Unseen categories (Retrieved template) | CD: 4.31 -> 3.69 | Improved |
| switch from retrieved to random template | KP-RED · Seen categories | CD: 3.05 -> 5.10 | Hurt |
| switch from retrieved to random template | Our-MV · Seen categories | CD: 2.38 -> 2.46 | No clear effect |
| replace TAX3D diffusion with single-step rectified flow matching | TAX3D · Unseen objects | Time (s): 3.23 -> 0.64 | 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.
- KP-RED — KP-RED: Exploiting Semantic Keypoints for Joint 3D Shape Retrieval and Deformation (CVPR 2024) · better on 6 of 6 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
- ShapeFlow: Learnable Deformation Flows Among 3D Shapes (NeurIPS 2020 · similarity 0.89)
- Topologically-Aware Deformation Fields for Single-View 3D Reconstruction (CVPR 2022 · similarity 0.88)
- Geometry-Aware Network for Non-Rigid Shape Prediction From a Single View (CVPR 2018 · similarity 0.88)
- LASR: Learning Articulated Shape Reconstruction From a Monocular Video (CVPR 2021 · similarity 0.88)
- Learning Monocular 3D Reconstruction of Articulated Categories From Motion (CVPR 2021 · similarity 0.88)
- Learning Category-Specific Deformable 3D Models for Object Reconstruction (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.
