Research map / 3D vision / Gaussian splatting
Rethinking Pose Refinement in 3D Gaussian Splatting under Pose Prior and Geometric Uncertainty
Mangyu Kong, Jaewon Lee, Seongwon Lee, Euntai Kim
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 Monte Carlo pose sampling and Fisher-information PnP to GS-CPR refinement | DFNet prior · 7Scenes | recall within [2cm, 2deg]: 76.5 -> 95.1 | Improved |
| use probabilistic UGS-Loc instead of iterating the deterministic GS-CPR twice (ablation) | DFNet prior · 7Scenes | recall within [2cm, 2deg]: 86.1 -> 95.1 | Improved |
| iterate deterministic GS-CPR a second time (ablation) | GS-CPR with ACE prior · 7Scenes | recall within [2cm, 2deg]: 93.1 -> 93.1 | No clear effect |
| iterate deterministic GS-CPR a second time (ablation) | GS-CPR with DFNet prior · 7Scenes | recall within [2cm, 2deg]: 76.5 -> 86.1 | Improved |
| add UGS-Loc to a weak pose prior | DFNet · Cambridge Landmarks (avg) | median translation error (cm) / rotation (deg): 26/0.51 -> 10.7/0.20 | Improved |
| increase number of particles from 2 to 16 (ablation) | UGS-Loc with DFNet prior · Cambridge Landmarks (avg) | median translation (cm) / rotation (deg) error: 11.8/0.24 -> 10.3/0.20 | Improved |
| increase number of particles from 2 to 16 (ablation) | UGS-Loc with ACE prior · Cambridge Landmarks (avg) | median translation (cm) / rotation (deg) error: 10.9/0.51 -> 9.6/0.20 | Improved |
| use MASt3R matcher instead of SuperPoint + LightGlue (ablation) | UGS-Loc with DFNet prior · Cambridge Landmarks (avg) | median translation (cm) / rotation (deg) error: 13/0.26 -> 11/0.20 | Improved |
| use MASt3R matcher instead of SuperPoint + LightGlue (ablation) | UGS-Loc with ACE prior · Cambridge Landmarks (avg) | median translation (cm) / rotation (deg) error: 11/0.26 -> 11/0.21 | No clear effect |
| swap in SuperPoint + LightGlue matcher in UGS-Loc versus GS-CPR baseline (ablation) | UGS-Loc with DFNet prior · Cambridge Landmarks (avg) | median translation (cm) / rotation (deg) error: 26/0.51 -> 13/0.26 | Improved |
| add UGS-Loc refinement to ACE prior on 12Scenes | ACE + GS-CPR · 12Scenes | recall within [2cm, 2deg]: 98.7 -> 99.1 | Improved |
| add UGS-Loc refinement to Marepo prior on 12Scenes | Marepo + GS-CPR · 12Scenes | recall within [2cm, 2deg]: 90.9 -> 95.3 | 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.
- GLACE — GLACE: Global Local Accelerated Coordinate Encoding (CVPR 2024) · better on 1 of 1 shared result
- DFNet — DFNet: Enhance Absolute Pose Regression with Direct Feature Matching (ECCV 2022) · better on 1 of 1 shared result
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
- GS-CPR: Efficient Camera Pose Refinement via 3D Gaussian Splatting (ICLR 2025 · similarity 0.90)
- DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization (NeurIPS 2024 · similarity 0.90)
- Metropolis-Hastings Sampling for 3D Gaussian Reconstruction (NeurIPS 2025 · similarity 0.90)
- MotionGS: Exploring Explicit Motion Guidance for Deformable 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.89)
- IndoorGS: Geometric Cues Guided Gaussian Splatting for Indoor Scene Reconstruction (CVPR 2025 · similarity 0.89)
- KeyGS: A Keyframe-Centric Gaussian Splatting Method for Monocular Image Sequences (AAAI 2025 · similarity 0.89)
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.
