Research map / 3D vision / Depth and stereo
Geometric Loss Functions for Camera Pose Regression With Deep Learning
Alex Kendall, Roberto Cipolla
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 |
|---|---|---|---|
| replace the fixed position-orientation weight with learned homoscedastic task uncertainty | PoseNet · Cambridge Landmarks, King's College | accuracy <2m, 5°: 65.0% -> 85.3% | Improved |
| replace the fixed position-orientation weight with learned homoscedastic task uncertainty | PoseNet · Dubrovnik 6K | accuracy <10m, 10°: 30.1% -> 41.7% | Improved |
| fine-tune a homoscedastic-uncertainty-pretrained model with reprojection loss | PoseNet · Cambridge Landmarks, King's College | accuracy <2m, 5°: 85.3% -> 90.3% | Improved |
| fine-tune a homoscedastic-uncertainty-pretrained model with reprojection loss | PoseNet · Dubrovnik 6K | accuracy <10m, 10°: 41.7% -> 48.6% | Improved |
| train with reprojection loss from random initialization rather than pretrained weights (ablation) | PoseNet · Cambridge Landmarks, King's College | convergence | Hurt |
| use robust Huber or Tukey regression norms instead of L1 (ablation) | PoseNet · datasets used in the paper | localization performance | Hurt |
| use reprojection loss instead of learned homoscedastic weighting (ablation) | PoseNet · Cambridge Landmarks and 7 Scenes | median position error: King's College: 0.99m -> 0.88m; Old Hospital: 2.17m -> 3.20m; Chess: 0.14m -> 0.13m | Mixed |
Closest papers by meaning
- Understanding the Limitations of CNN-Based Absolute Camera Pose Regression (CVPR 2019 · similarity 0.89)
- Wide-Baseline Relative Camera Pose Estimation With Directional Learning (CVPR 2021 · similarity 0.89)
- Neural Reprojection Error: Merging Feature Learning and Camera Pose Estimation (CVPR 2021 · similarity 0.89)
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.88)
- Geometry-Aware Learning of Maps for Camera Localization (CVPR 2018 · similarity 0.88)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · 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.
