Research map / 3D vision / Human pose estimation
Are Minimal Radial Distortion Solvers Really Necessary for Relative Pose Estimation?
Viktor Kocur, Charalambos Tzamos, Yaqing Ding, Zuzana Berger Haladová, Torsten Sattler, Zuzana Kúkelová
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 sampled undistortion parameters {0.0,-0.6,-1.2} instead of lambda=0 (ablation) | 6pt Ef solver · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 15.13 -> 13.86 | Improved |
| add sampled undistortion parameters instead of lambda=0 (ablation) | 6pt Ef solver · PragueParks synthetic A, shared intrinsics | avg pose error (deg): 11.48 -> 7.45 | Improved |
| add sampled undistortion parameters instead of lambda=0 (with lambda refinement) (ablation) | 7pt F solver · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 43.13 -> 35.36 | Improved |
| refine distortion and focal parameters in local optimization (ablation) | 7pt F solver, lambda=0 · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 50.51 -> 43.13 | Improved |
| refine lambda in local optimization after GeoCalib prediction (ablation) | 6pt Ef + GeoCalib · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 15.63 -> 14.45 | Improved |
| refine all camera parameters in LO after GeoCalib prediction (ablation) | 5pt E + GeoCalib · ETH3D synthetic A, different cameras | avg pose error (deg): 33.15 -> 29.44 | Improved |
| refine all camera parameters in LO after GeoCalib prediction (ablation) | 5pt E + GeoCalib · PragueParks synthetic B, different cameras | avg pose error (deg): 48.20 -> 41.94 | Improved |
| refine focal length and lambda in LO after GeoCalib prediction (ablation) | 3pt E + GeoCalib · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 22.12 -> 25.00 | Hurt |
| use non-minimal 10pt solver instead of minimal 9pt solver | F-lambda1-lambda2 · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 42.46 -> 38.45 | Improved |
| use 6pt Ef with sampled lambda instead of minimal 8pt F-lambda solver | 6pt Ef · ETH3D synthetic A, shared intrinsics | avg pose error (deg): 31.78 -> 13.86 | Improved |
| use 6pt Ef with sampled lambda instead of minimal 9pt F-lambda1-lambda2 solver | 6pt Ef · ETH3D synthetic A, shared intrinsics | RANSAC time (ms): 826.20 -> 142.94 | Improved |
| use 3pt E with gravity prior instead of GeoCalib 5pt on natural scenes | 3pt E + GeoCalib · PragueParks synthetic A, shared intrinsics | avg pose error (deg): 21.36 -> 46.57 | Hurt |
Closest papers by meaning
- Revisiting Radial Distortion Absolute Pose (ICCV 2019 · similarity 0.94)
- Camera Pose Estimation With Unknown Principal Point (CVPR 2018 · similarity 0.88)
- Radial Distortion Invariant Factorization for Structure From Motion (ICCV 2021 · similarity 0.87)
- PRaDA: Projective Radial Distortion Averaging (CVPR 2025 · similarity 0.86)
- From Correspondences to Pose: Non-minimal Certifiably Optimal Relative Pose without Disambiguation (CVPR 2024 · similarity 0.85)
- Map-Relative Pose Regression for Visual Re-Localization (CVPR 2024 · 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.
