Research map / Continual learning / Online and replay-based
Continual Learning for Image-Based Camera Localization
Shuzhe Wang, Zakaria Laskar, Iaroslav Melekhov, Xiaotian Li, Juho Kannala
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 coverage-score buffering instead of class-balance buffering | HSCNet continual learning · i7S buffer 256 Img-buff | pose accuracy %: 70.06 -> 72 | Improved |
| add coverage-score buffering instead of reservoir sampling | HSCNet continual learning · i12S buffer 256 Img-buff | pose accuracy %: 82.54 -> 91.85 | Improved |
| add coverage-score buffering instead of class-balance buffering | HSCNet continual learning · i19S buffer 256 Img-buff | pose accuracy %: 64.7 -> 68 | Improved |
| add coverage-score buffering instead of class-balance buffering | HSCNet continual learning · i19S buffer 128 Img-buff | pose accuracy %: 47.98 -> 46.43 | Hurt |
| add coverage-score buffering instead of class-balance buffering, 5-seed runs | HSCNet continual learning · i7S buffer 256 Img-buff | final accuracy %: 69.66 ± 0.54 -> 70.51 ± 0.94 | Improved |
| enlarge buffer from 128 to 1024 for Buff-CS coverage score on i19S (ablation) | Buff-CS · i19S | coverage score %: 58.8 -> 96 | Improved |
| enlarge buffer from 128 to 256 for Buff-CS (ablation) | Buff-CS · i19S Img-buff | pose accuracy %: 46.43 -> 68 | Improved |
| store intermediate representations (Rep-buff) in addition to images (ablation) | Buff-CS · i19S buffer 128 | pose accuracy %: 46.43 -> 51.86 | Improved |
| store intermediate representations (Rep-buff) instead of images only (ablation) | Class-balance · i19S buffer 128 | pose accuracy %: 47.98 -> 54.12 | Improved |
| store intermediate representations (Rep-buff) instead of images only (ablation) | Buff-CS · i7S buffer 256 | pose accuracy %: 72 -> 71.34 | No clear effect |
| add any replay buffer versus training without buffering (ablation) | HSCNet continual learning · i7S buffer 256 | pose accuracy %: 10.61 -> 72.00 | Improved |
| Rep-buff storage compared to Img-buff (ablation) | Buff-CS · buffer 256 memory | storage (Mb): 117 -> 1091 | Hurt |
Closest papers by meaning
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.87)
- Learn to Memorize and to Forget: A Continual Learning Perspective of Dynamic SLAM (ECCV 2024 · similarity 0.86)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.85)
- R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization (CVPR 2025 · similarity 0.84)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.84)
- ConDo: Continual Domain Expansion for Absolute Pose Regression (AAAI 2025 · 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.
