Research map / 3D vision / Depth and stereo
Boosting 3-DoF Ground-to-Satellite Camera Localization Accuracy via Geometry-Guided Cross-View Transformer
Yujiao Shi, Fei Wu, Akhil Perincherry, Ankit Vora, Hongdong Li
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 |
|---|---|---|---|
| replace joint LM optimization with ground-to-satellite synthesis plus neural pose optimizer and dense search | cross-view camera localization · KITTI Test1 with aligned orientation | Lateral rate d=1 (%): 64.27 -> 93.85 | Improved |
| use neural pose optimizer for rotation instead of deep LM | decoupled localization pipeline · KITTI Test1, 20 deg noise | Azimuth rate theta=1 (%): 19.64 -> 99.10 | Improved |
| estimate translation with neural optimizer instead of LM (ablation) | neural pose optimizer (R & t) · KITTI Test1, 20 deg noise | Lateral rate d=1 (%): 27.72 -> 4.90 | Hurt |
| replace LM translation with dense search spatial correlation (ablation) | Neural Opt. (R) + translation estimator · KITTI Test1, 20 deg noise | Lateral rate d=1 (%): 41.82 -> 64.83 | Improved |
| add geometry-guided cross-view transformer over geometry projection (ablation) | Neural Opt. (R) + DS (t) without uncertainty · KITTI Test1, 20 deg noise | Lateral rate d=1 (%): 64.83 -> 70.29 | Improved |
| add uncertainty map to spatial correlation (ablation) | full method · KITTI Test1, 20 deg noise | Lateral rate d=1 (%): 70.29 -> 76.44 | Improved |
| geometry-guided transformer instead of Persformer for G2S synthesis (ablation) | full method · KITTI Test1, 20 deg noise | Longitudinal rate d=1 (%): 5.51 -> 23.54 | Improved |
| add translation supervision on neural optimizer output (ablation) | neural pose optimizer · KITTI Test1, 20 deg noise | Azimuth rate theta=1 (%): 10.52 -> 99.10 | Improved |
| use ground-to-satellite synthesis instead of satellite-to-ground (ablation) | full method · KITTI Test1, 80 deg noise | Azimuth rate theta=1 (%): 19.56 -> 53.27 | Improved |
| increase neural optimizer iterations from 2 to 3 (ablation) | neural pose optimizer · KITTI Test1 | Azimuth rate theta=1 (%): 99.10 -> 80.52 | No clear effect |
| compare dense-search method with CVML regression on mean error | full method · Oxford RobotCar overall | Mean error (m): 2.07 -> 2.79 | Hurt |
| dense-search similarity matching versus CVML regression on median error | full method · Oxford RobotCar overall | Median error (m): 1.44 -> 1.02 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SliceMatch — SliceMatch: Geometry-Guided Aggregation for Cross-View Pose Estimation (CVPR 2023) · better on 2 of 2 shared results
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
- Beyond Cross-View Image Retrieval: Highly Accurate Vehicle Localization Using Satellite Image (CVPR 2022 · similarity 0.93)
- VIRD: View-Invariant Representation through Dual-Axis Transformation for Cross-View Pose Estimation (CVPR 2026 · similarity 0.91)
- Convolutional Cross-View Pose Estimation (TPAMI 2023 · similarity 0.91)
- BevSplat: Resolving Height Ambiguity via Feature-Based Gaussian Primitives for Weakly-Supervised Cross-View Localization (NeurIPS 2025 · similarity 0.90)
- View From Above: Orthogonal-View aware Cross-view Localization (CVPR 2024 · similarity 0.89)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.89)
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.
