Research map / 3D vision / Depth and stereo
Coarse-to-Fine Multi-Scene Pose Regression With Transformers
Yoli Shavit, Ron Ferens, Yosi Keller
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 coarse-to-fine classification-regression with centroid residuals and fine-tune from MS-Transformer init | MS-Transformer · Cambridge Landmarks | avg position error (m): 1.28 -> 0.98 | Improved |
| initialize from pretrained MS-Transformer (ablation) | c2f-MS-Transformer · Cambridge Landmarks | avg position error (m): 1.27 -> 0.98 | Improved |
| initialize from pretrained MS-Transformer (ablation) | c2f-MS-Transformer · 7Scenes | avg position error (m): 0.18 -> 0.17 | No clear effect |
| use EfficientNetB0 instead of ResNet50 backbone (ablation) | c2f-MS-Transformer · 7Scenes | position/orientation error: 0.19 -> 0.18 m; 8.60 -> 7.55 deg | Improved |
| use EfficientNetB1 instead of EfficientNetB0 (ablation) | c2f-MS-Transformer · 7Scenes | position/orientation error: 0.18 -> 0.17 m; 7.55 -> 7.26 deg | Improved |
| use coarse position map 14x14x112 with fine orientation map 28x28x40 instead of fine maps for both (ablation) | c2f-MS-Transformer · 7Scenes | avg position error (m): 0.22 -> 0.18 | Improved |
| use 7x7x320 position map instead of 14x14x112 (ablation) | c2f-MS-Transformer · 7Scenes | avg position error (m): 0.18 -> 0.19 | Hurt |
| use coarse orientation map 14x14x112 instead of 28x28x40 (ablation) | c2f-MS-Transformer · 7Scenes | avg orientation error (deg): 7.55 -> 7.96 | Hurt |
| increase encoder/decoder layers from 2 to 6 (ablation) | c2f-MS-Transformer · 7Scenes | avg orientation error (deg): 8.21 -> 7.55 | Improved |
| increase transformer dimension from 64 to 512 (ablation) | c2f-MS-Transformer · 7Scenes | avg orientation error (deg): 8.32 -> 7.21 | Improved |
| use 4 position/orientation clusters instead of 8 (ablation) | c2f-MS-Transformer · Cambridge Landmarks | avg position error (m): 1.69 -> 1.27 | Improved |
| use a single position cluster instead of more clusters (ablation) | c2f-MS-Transformer · 7Scenes | avg position error (m): 0.25 -> 0.18 | Improved |
Closest papers by meaning
- Map-Relative Pose Regression for Visual Re-Localization (CVPR 2024 · similarity 0.89)
- ConDo: Continual Domain Expansion for Absolute Pose Regression (AAAI 2025 · similarity 0.87)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.87)
- Neural Refinement for Absolute Pose Regression with Feature Synthesis (CVPR 2024 · similarity 0.86)
- Scene-agnostic Pose Regression for Visual Localization (CVPR 2025 · similarity 0.86)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · 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.
