Research map / 3D vision / Depth and stereo
Accelerated Coordinate Encoding: Learning to Relocalize in Minutes Using RGB and Poses
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| train a scene-specific MLP head on randomly mixed features from many mapping views rather than training on one image at a time | ACE scene coordinate regression · 7Scenes and 12Scenes | mapping time: 15h -> 5 min | Helped |
| replace end-to-end training through a robust pose solver with a curriculum over reprojection loss (ablation) | ACE scene coordinate regression · 7Scenes | localized frames (<5cm, 5°): 94.9 -> 95.3 | Helped |
| use a fixed 50px tanh threshold instead of the circular schedule with homogeneous coordinates (ablation) | ACE · 7Scenes | localized frames (<5cm, 5°): 97.1 -> 94.8 | Hurt |
| use a linear tanh-threshold schedule instead of the circular schedule with homogeneous coordinates (ablation) | ACE · 7Scenes | localized frames (<5cm, 5°): 97.1 -> 94.6 | Hurt |
| add homogeneous coordinate prediction to the circular tanh schedule (ablation) | ACE · 7Scenes | localized frames (<5cm, 5°): 95.3 -> 97.1 | Helped |
| use a four-model ensemble of independently trained ACE heads over spatially clustered mapping frames (ablation) | ACE · Cambridge Landmarks, average | median position error (cm) / rotation error (°): 25/0.4 -> 17/0.3 | Helped |
| increase the ACE ensemble from four to seven models (ablation) | ACE ensemble · Cambridge Landmarks, average | median position error (cm) / rotation error (°): 17/0.3 -> 16/0.3 | Helped |
| increase the ACE ensemble from four to seven models (ablation) | ACE ensemble · Cambridge Landmarks, mapping time | mapping time: 20min -> 35min | Hurt |
| use 5.5MB rather than the standard 4MB ACE map (ablation) | ACE · 7Scenes | localized frames | Hurt |
| use 2.5MB rather than the standard 4MB ACE map (ablation) | ACE · 7Scenes | localized frames | Hurt |
Closest papers by meaning
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.89)
- R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization (CVPR 2025 · similarity 0.87)
- Living Scenes: Multi-object Relocalization and Reconstruction in Changing 3D Environments (CVPR 2024 · similarity 0.86)
- Map-Relative Pose Regression for Visual Re-Localization (CVPR 2024 · similarity 0.86)
- Change-Robust Online Topological Memory for Long-Term Relocalization and Semantic Navigation (NeurIPS 2026 · similarity 0.86)
- GLACE: Global Local Accelerated Coordinate Encoding (CVPR 2024 · similarity 0.86)
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.