Research map / 3D vision / Depth and stereo
Semantic-guided Camera Ray Regression for Visual Localization
Yesheng Zhang, Xu Zhao
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 |
|---|---|---|---|
| regress camera rays instead of scene coordinates in the same architecture (ablation) | DIMM · MapFree s00000-s00005 | recall at 5cm/5deg: 72.6 -> 87.1 | Improved |
| regress absolute camera pose instead of rays (ablation) | DIMM · MapFree s00000-s00005 | recall at 5cm/5deg: 87.1 -> 4.6 | Hurt |
| add CLS token and XY position tokens to patch features (ablation) | DIMM feature encoder · MapFree s00006 | recall at 5cm/5deg: 17.2 -> 44.3 | Improved |
| use Fourier encoding of patch position instead of raw XY (ablation) | DIMM feature encoder · MapFree s00006 | recall at 5cm/5deg: 27.2 -> 44.3 | Improved |
| fine-tune last two DINO blocks (ablation) | DIMM feature encoder · MapFree s00006 | recall at 5cm/5deg: 39.7 -> 44.3 | Improved |
| use DINO CLS token instead of image-retrieval global feature (ablation) | DIMM feature encoder · MapFree s00006 | recall at 5cm/5deg: 19.7 -> 44.3 | Improved |
| add semantic attention layers over averaging sub-mappers (ablation) | DIMM mapper head · MapFree s00006 | recall at 5cm/5deg: 31.0 -> 44.3 | Improved |
| increase semantic attention layers from 4 to 8 (ablation) | DIMM mapper head · MapFree s00006 | recall at 5cm/5deg: 44.3 -> 30.4 | Hurt |
| add RC-RANSAC for pose estimation (ablation) | DIMM · MapFree s00006 | recall at 5cm/5deg: 44.3 -> 64.8 | Improved |
| add RC-RANSAC on Indoor-6 scene2a (ablation) | DIMM · Indoor-6 scene2a | recall at 5cm/5deg: 66.7 -> 77.4 | Improved |
| camera ray regression versus ACE on rotation accuracy | DIMM · MapFree | rotation error | Hurt |
| RC-RANSAC adds computation at inference | DIMM-R · MapFree | throughput (fps): 57.7 -> 20.1 | Hurt |
Closest papers by meaning
- ACE-G: Improving Generalization of Scene Coordinate Regression Through Query Pre-Training (ICCV 2025 · similarity 0.87)
- R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization (CVPR 2025 · similarity 0.86)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.86)
- Where Am I and What Will I See: An Auto-Regressive Model for Spatial Localization and View Prediction (ICLR 2025 · similarity 0.85)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.85)
- GLACE: Global Local Accelerated Coordinate Encoding (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.
