Research map / 3D vision / Depth and stereo
GeoMix: Descriptor-Free Visual Localization via Global Context and Multi-Detector Training
Yejun zhang, Xinjue Wang, Zihan Wang, Esa Rahtu, 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 learnable global context nodes (ablation) | A2-GNN · MegaDepth | Reproj. AUC@5px (%): 52.87 -> 54.59 | Improved |
| add directional and distance embeddings to annular conv branch (ablation) | A2-GNN + GC nodes · MegaDepth | Reproj. AUC@5px (%): 54.59 -> 55.53 | Improved |
| add geometry embedding to maxpooling branch (ablation) | A2-GNN + GC + annular · MegaDepth | Reproj. AUC@5px (%): 55.53 -> 58.34 | Improved |
| add Mix-Training over SIFT, SuperPoint, DISK (ablation) | GeoMix without Mix-Training · MegaDepth | Reproj. AUC@5px (%): 58.34 -> 65.61 | Improved |
| add Mix-Training (ablation) | GeoMix · MegaDepth | Rotation error 75% quantile (deg): 2.26 -> 0.52 | Improved |
| Mix-Training instead of SIFT-only training, tested on unseen R2D2 (ablation) | GeoMix · MegaDepth (R2D2) | median rotation error (deg): 1.94 -> 0.34 | Improved |
| Mix-Training instead of SIFT-only training, tested on unseen DeDoDe v2 (ablation) | GeoMix · MegaDepth (DeDoDe v2) | median rotation error (deg): 0.39 -> 0.17 | Improved |
| Mix-Training instead of SIFT-only training, tested on SIFT (ablation) | GeoMix · MegaDepth (SIFT) | Reproj. AUC@5px (%): 58.34 -> 65.61 | Improved |
| Mix-Training instead of SIFT-only training, tested on SuperPoint (ablation) | GeoMix · MegaDepth (SuperPoint) | Reproj. AUC@5px (%): 8.55 -> 20.84 | Improved |
| GeoMix instead of A2-GNN | descriptor-free matcher · MegaDepth | Rotation error 75% quantile (deg): 4.60 -> 0.52 | Improved |
| GeoMix instead of A2-GNN on Old Hospital | descriptor-free matcher · Cambridge Landmarks | translation error (cm): 59 -> 27 | Improved |
| GeoMix instead of A2-GNN on Stairs | descriptor-free matcher · 7Scenes | translation / rotation error: 72 / 17.0 -> 45 / 11.1 | Improved |
Closest papers by meaning
- Is Geometry Enough for Matching in Visual Localization? (ECCV 2022 · similarity 0.91)
- Towards Visual Localization Interoperability: Cross-Feature for Collaborative Visual Localization and Mapping (ICCV 2025 · similarity 0.88)
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.88)
- SemLight: Distilled Semantic–Geometric Fusion for Efficient Local Feature Matching (ECCV 2026 · similarity 0.88)
- Fully Geometric Panoramic Localization (CVPR 2024 · similarity 0.88)
- GeoComplete: Geometry-Aware Diffusion for Reference-Driven Image Completion (NeurIPS 2025 · similarity 0.87)
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.
