| Full name | Image retrieval using late fusion of img. descs. and local R2D2 features |
| Description | Image retrieval using late fusion of global image descriptors (top-50, mapping and query) and local R2D2 (20k) features.
3D reconstruction and image registration is done with COLMAP. |
| Publication title | Robust Image Retrieval-based Visual Localization using Kapture |
| Publication authors | Martin Humenberger, Yohann Cabon, Nicolas Guerin, Julien Morat, Jérôme Revaud, Philippe Rerole, Noé Pion, Cesar de Souza, Vincent Leroy, Gabriela Csurka |
| Publication URL | https://arxiv.org/abs/2007.13867 |
| Input Data Types | Uses Color Uses 2D |
| Programming language(s) | Python |
| Hardware | GPU (Tesla v100) |
| Website | https://europe.naverlabs.com/research/3d-vision/kapture |
| Source code or download URL | https://github.com/naver/kapture-localization |
| Submission creation date | 18 Jun, 2021 |
| Last edited | 6 Oct, 2021 |
| Info | Score | DCRE (0.05) | DCRE (0.15) | Pose (0.05m, 5°) | Outlier (0.5) | NaN or N/A |
|---|
 | 1.338 | 0.447 | 0.612 | 0.176 | 0.109 | 0.000 |