Research map / 3D vision / Depth and stereo
Loc$^{2}$: Interpretable Cross-View Localization via Depth-Lifted Local Feature Matching
Zimin Xia, Chenghao Xu, Alexandre Alahi
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 |
|---|---|---|---|
| make Procrustes alignment scale-aware (estimate depth scale) (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 5.47 -> 3.86 | Improved |
| use all matched points instead of only topmost points (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 3.95 -> 3.86 | No clear effect |
| add RANSAC at inference (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 4.60 -> 3.86 | Improved |
| match on the image plane instead of a BEV-transformed ground image (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 9.49 -> 3.86 | Improved |
| use scale-aware Procrustes instead of homography pose estimation (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 14.46 -> 3.86 | Improved |
| set matching temperature to 0.1 instead of 0.5 or 0.01 (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 9.58 -> 3.86 (tau 0.5 vs 0.1); 4.58 (tau 0.01) | Improved |
| sample 1024 correspondences in training instead of 256 (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 4.31 -> 3.86 | Improved |
| increase aerial points beyond 41x41 (up to 56x56) (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 3.86 -> 3.70 | No clear effect |
| upsample the aerial DINO feature map before sampling points (ablation) | Loc2 · VIGOR same-area validation (unknown orientation) | mean localization error (m): 3.86 -> 4.55 | Hurt |
| replace metric depth with relative depth (BiFuse++/UniFuse) at inference (ablation) | Loc2 · VIGOR cross-area (unknown orientation) | mean localization error (m): 4.23 -> 4.43 (BiFuse++), 4.36 (UniFuse) | No clear effect |
| match local features on image plane with depth lifting vs FG2 BEV matching, unknown orientation | Loc2 vs FG2 · VIGOR cross-area (unknown orientation) | mean localization error (m): 10.02 -> 4.23 | Improved |
| Loc2 versus FG2 when orientation is known | Loc2 vs FG2 · VIGOR cross-area (known orientation) | mean localization error (m): 2.41 -> 3.43 | Hurt |
Later papers that compared with this method
2 later accepted papers report Loc2 as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- ARC-Loc: Leveraging Azimuthal Ray Convergence as a Geometric Cue for Direct Cross-View Localization (ECCV 2026 · better on 3 of 4 shared results)
- Georeferenced Cross-View 3D Reconstruction from Ground and Satellite Images (NeurIPS 2026 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SliceMatch — SliceMatch: Geometry-Guided Aggregation for Cross-View Pose Estimation (CVPR 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FG^2: Fine-Grained Cross-View Localization by Fine-Grained Feature Matching (CVPR 2025 · similarity 0.92)
- View From Above: Orthogonal-View aware Cross-view Localization (CVPR 2024 · similarity 0.90)
- BevSplat: Resolving Height Ambiguity via Feature-Based Gaussian Primitives for Weakly-Supervised Cross-View Localization (NeurIPS 2025 · similarity 0.89)
- From Coarse to Fine: A Matching and Alignment Framework for Unsupervised Cross-View Geo-Localization (AAAI 2025 · similarity 0.88)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.88)
- LoD-Loc: Aerial Visual Localization using LoD 3D Map with Neural Wireframe Alignment (NeurIPS 2024 · 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.
