Research map / 3D vision / Human pose estimation
Wide-Baseline Relative Camera Pose Estimation With Directional Learning
Kefan Chen, Noah Snavely, Ameesh Makadia
Check 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 |
|---|---|---|---|
| Represent pose components as discrete directional distributions rather than directly regressing pose | DirectionNet versus 6D regression · Matterport-A, Matterport-B, InteriorNet-A, InteriorNet-B | rotation and translation mean angular error: Matterport-A rotation 5.73 -> 4.30; translation 35.75 -> 16.37; Matterport-B rotation 18.23 -> 14.85; translation 39.06 | Improved |
| Factorize rotation into spherical direction vectors instead of estimating a discrete quaternion distribution | DirectionNet · Matterport-A and Matterport-B | rotation mean angular error: 3.96 vs 23.32; 13.60 vs 37.09 | Improved |
| Add intermediate derotation and estimate translation in a separate second stage (ablation) | DirectionNet-9D · Matterport-A, Matterport-B, InteriorNet-A, InteriorNet-B | translation mean angular error: 22.84 -> 14.17; 29.45 -> 21.26; 19.53 -> 12.36; 31.87 -> 16.36 | Improved |
| Use SVD projection rather than partial Gram-Schmidt projection (ablation) | DirectionNet rotation estimation · Matterport-A and InteriorNet-A | rotation mean angular error: 4.30 -> 3.96; 2.90 -> 2.87 | Improved |
| Add distribution loss (ablation) | DirectionNet-9D · Matterport-A | rotation mean angular error: 14.66 -> 3.96 | Improved |
| Add direction loss (ablation) | DirectionNet-9D · Matterport-A | rotation mean angular error: 4.85 -> 3.96 | Improved |
| Add spread loss (ablation) | DirectionNet-9D · Matterport-A | rotation mean angular error: 4.68 -> 3.96 | Improved |
| Perturb the estimated rotation used for derotation during translation training (ablation) | DirectionNet-T · InteriorNet and Matterport-B | translation angular error: 4° worse on InteriorNet without perturbation | Improved |
| Use dense MSE supervision for distributions instead of KL divergence (ablation) | DirectionNet · Experiments reported in paper | performance | Improved |
| Train with 100% rather than 20% of KITTI training data (ablation) | DirectionNet-9D · KITTI | rotation and translation mean angular error: 10.50 -> 9.19; 26.74 -> 19.36 | Improved |
| Use SIFT+LMedS rather than SIFT+RANSAC (ablation) | SIFT feature-based pose estimation · Matterport and InteriorNet datasets | rotation and translation angular error | Mixed |
Closest papers by meaning
- Fusing the Old with the New: Learning Relative Camera Pose with Geometry-Guided Uncertainty (CVPR 2021 · similarity 0.90)
- Neural Reprojection Error: Merging Feature Learning and Camera Pose Estimation (CVPR 2021 · similarity 0.89)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.89)
- FAR: Flexible Accurate and Robust 6DoF Relative Camera Pose Estimation (CVPR 2024 · similarity 0.89)
- DVMNet: Computing Relative Pose for Unseen Objects Beyond Hypotheses (CVPR 2024 · similarity 0.88)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · 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.
