Research map / 3D vision / Depth and stereo
HOLO: Homography-Guided Pose Estimator Network for Fine-Grained Visual Localization on SD Maps
Xuchang Zhong, Xu Cao, Jinke Feng, Hao Fang
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 3-DoF pose regression head with homography head (ablation) | HOLO-CA · nuScenes val | Position Recall@1m: 10.06 -> 21.47 | Improved |
| replace 3-DoF pose regression head with homography head (ablation) | HOLO (one iteration) · nuScenes val | Position Recall@1m: 9.88 -> 26.57 | Improved |
| homography head on orientation (ablation) | HOLO (one iteration) · nuScenes val | AOE (deg): 2.19 -> 1.49 | Improved |
| increase refinement iterations from 1 to 6 (ablation) | HOLO · nuScenes val | Position Recall@1m: 26.57 -> 36.41 | Improved |
| increase iterations from 2 to 6 on coarse metrics (ablation) | HOLO · nuScenes val | Orientation Recall@5deg: 98.14 -> 97.35 | No clear effect |
| homography-guided fusion instead of cross-attention fusion (ablation) | HOLO · nuScenes val | Position Recall@1m: 21.47 -> 26.57 | Improved |
| add joint homography learning to BEV perception training (ablation) | BEV segmentation · nuScenes val | IoU drivable area: 76.52 -> 79.89 | Improved |
| add joint homography learning to BEV perception training (ablation) | BEV segmentation · nuScenes val | IoU buildings: 68.41 -> 69.43 | Improved |
| use coarser BEV grid 0.5 mpp instead of 0.25 (ablation) | HOLO · nuScenes val | Position Recall@1m: 36.41 -> 36.10 | No clear effect |
| increase simulated map noise from 30 m to 40 m (ablation) | HOLO · nuScenes val | Position Recall@1m: 36.41 -> 18.24 | Hurt |
| supervise with SD road layer instead of HD drivable layer (ablation) | HOLO · nuScenes val | Position Recall@1m: 36.41 -> 30.40 | Hurt |
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.
- OrienterNet — OrienterNet: Visual Localization in 2D Public Maps With Neural Matching (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
- Deep Homography Estimation for Visual Place Recognition (AAAI 2024 · similarity 0.87)
- Geometry-Aware Learning of Maps for Camera Localization (CVPR 2018 · similarity 0.86)
- Long-Term Visual Localization Revisited (TPAMI 2020 · similarity 0.86)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.86)
- View Consistent Purification for Accurate Cross-View Localization (ICCV 2023 · similarity 0.85)
- A Guide to Structureless Visual Localization (IJCV 2026 · similarity 0.85)
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.
