Research map / 3D vision / Human pose estimation
G2FM: A Geodesic Flow Matching Framework with Geometric Prior for Category-Level 9-DoF Pose Estimation
Qianyu Chen, Xiaogang Zhang, Yangyi Wan, Zhengzhao Pan, Kai Wang, Yuqi Cai, Wenbin Yan, feng yang, Hua Chen
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 |
|---|---|---|---|
| replace Euclidean flow with SO(3) geodesic path and update (ablation) | Standard FM · Omni6DPose ROPE zero-shot (trained on CAMERA25) | 3D50: 0.12 -> 5.75 | Improved |
| add geometric guidance to SO(3) flow matching (ablation) | FM + SO(3) · Omni6DPose ROPE zero-shot (trained on CAMERA25) | 3D50: 5.75 -> 18.94 | Improved |
| add geometric guidance to SO(3) flow matching (ablation) | FM + SO(3) · Omni6DPose ROPE zero-shot (trained on CAMERA25) | 3D25: 40.16 -> 52.91 | Improved |
| use SO(3) geodesic update instead of Euclidean 6D flow (ablation) | Euclidean 6D-FM · REAL275 rotation-path ablation | 3D50: 76.6 -> 79.4 | Improved |
| use SO(3) geodesic update instead of Euclidean 6D flow (ablation) | Euclidean 6D-FM · REAL275 rotation-path ablation | 5deg5cm: 44.5 -> 39.2 | Hurt |
| use geodesic update instead of matrix-FM with SVD projection (ablation) | Matrix-FM + SVD · REAL275 rotation-path ablation | 3D75: 21.3 -> 42.6 | Improved |
| replace ResNet18 with DINOv2 RGB encoder (ablation) | G2FM · REAL275 encoder ablation | 3D50: 80.0 -> 74.9 | Hurt |
| replace ResNet18 with SigLIP2 RGB encoder (ablation) | G2FM · REAL275 encoder ablation | 3D50: 80.0 -> 74.9 | Hurt |
| replace PointNet with DGCNN (ablation) | G2FM · REAL275 encoder ablation | 3D50: 80.0 -> 61.2 | Hurt |
| replace PointNet with DGCNN (ablation) | G2FM · REAL275 encoder ablation | 5deg2cm: 36.5 -> 35.0 | No clear effect |
| add training-free geometric guidance to Diff9D | Diff9D · Omni6DPose | 3D25: 0.72 -> 8.45 | Improved |
| add training-free geometric guidance to Diff9D | Diff9D · REAL275 | 3D50: 76.5 -> 77.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Diff9D — Diff9D: Diffusion-Based Domain-Generalized Category-Level 9-DoF Object Pose Estimation (TPAMI 2025) · better on 9 of 12 shared results
- SecondPose — SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation (CVPR 2024) · better on 2 of 3 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
- GeoReF: Geometric Alignment Across Shape Variation for Category-level Object Pose Refinement (CVPR 2024 · similarity 0.87)
- Vision Foundation Model Enables Generalizable Object Pose Estimation (NeurIPS 2024 · similarity 0.87)
- Metric Flow Matching for Smooth Interpolations on the Data Manifold (NeurIPS 2024 · similarity 0.87)
- NRDF: Neural Riemannian Distance Fields for Learning Articulated Pose Priors (CVPR 2024 · similarity 0.86)
- Direct Product Flow Matching: Decoupling Radial and Angular Dynamics for Few-Shot Adaptation (NeurIPS 2026 · similarity 0.86)
- FlowRAM: Grounding Flow Matching Policy with Region-Aware Mamba Framework for Robotic Manipulation (CVPR 2025 · similarity 0.86)
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.
