Research map / 3D vision / Human pose estimation
Two-View Geometry Scoring Without Correspondences
Axel Barroso-Laguna, Eric Brachmann, Victor Adrian Prisacariu, Gabriel J. Brostow, Daniyar Turmukhambetov
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 |
|---|---|---|---|
| Score pose hypotheses with FSNet instead of MAGSAC++ | Fundamental matrix estimation · ScanNet test, 0-100 correspondences | mAA translation: 0.12 -> 0.19 | Improved |
| Score pose hypotheses with FSNet instead of MAGSAC++ | Essential matrix estimation · ScanNet test, 0-100 correspondences | mAA max(R,t): 0.14 -> 0.18 | Improved |
| Combine MAGSAC++ candidate selection with FSNet reranking | Fundamental matrix estimation · ScanNet test, All | mAA max(R,t): 0.14 -> 0.19 | Improved |
| Combine MAGSAC++ candidate selection with FSNet reranking | Essential matrix estimation · PhotoTourism test, All | mAA translation: 0.65 -> 0.67 | Improved |
| Train using both fundamental- and essential-matrix hypotheses (ablation) | FSNet · ScanNet test, fundamental estimation, All | mAA translation: 0.21 -> 0.20 | Hurt |
| Train using soft-clamped pose-error regression rather than binary cross-entropy classification (ablation) | FSNet · ScanNet test, fundamental estimation | mAA translation, 0-100: 0.18 -> 0.19 | Improved |
| Add soft clamping to the L1 pose-error loss (ablation) | FSNet · ScanNet test, fundamental estimation | mAA translation, 0-100: 0.16 -> 0.19 | Improved |
| Supervise with pose error rather than Sampson distance (ablation) | FSNet · ScanNet test, fundamental estimation | mAA rotation, 0-100: 0.23 -> 0.29 | Improved |
| Use a correspondence-count filter to select between FSNet and MAGSAC++ (ablation) | Fundamental matrix estimation · ScanNet test, All | mAA max(R,t): 0.15 -> 0.16 | Improved |
| Loosen SuperGlue filtering to provide more correspondences (ablation) | MAGSAC++ · ScanNet test, fundamental estimation | mAA max(R,t): 0.14 -> 0.14 | No clear effect |
Closest papers by meaning
- Learning Two-View Correspondences and Geometry Using Order-Aware Network (ICCV 2019 · similarity 0.88)
- OANet: Learning Two-View Correspondences and Geometry Using Order-Aware Network (TPAMI 2020 · similarity 0.87)
- Minimal Solutions for Relative Pose With a Single Affine Correspondence (CVPR 2020 · similarity 0.87)
- On the Instability of Relative Pose Estimation and RANSAC's Role (CVPR 2022 · similarity 0.87)
- Condition Numbers in Multiview Geometry, Instability in Relative Pose Estimation, and RANSAC (TPAMI 2025 · similarity 0.87)
- BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024 · similarity 0.85)
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.
