Research map / 3D vision / Depth and stereo
NeRF-IBVS: Visual Servo Based on NeRF for Visual Localization and Navigation
Yuanze Wang, Yichao Yan, Dianxi Shi, Wenhan Zhu, Jianqiang Xia, Tan Jeff, Songchang Jin, KE GAO, XIAOBO LI, Xiaokang Yang
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 IBVS pose optimization after coarse pose estimation (ablation) | NeRF-IBVS · 12-Scenes average | median position/orientation error (m/deg): 0.27/13.10 -> 0.02/0.95 | Improved |
| add correspondence selection (threshold plus RANSAC) to IBVS (ablation) | NeRF-IBVS · 12-Scenes average | median position/orientation error (m/deg): 0.09/3.73 -> 0.02/0.95 | Improved |
| coarse pose estimation as starting point for optimization (frequency coarse error is reduced after optimization) (ablation) | NeRF-IBVS · 12-Scenes average | frequency of error reduction: 95.66% | Improved |
| lowest per-scene frequency of error reduction (Office5a) (ablation) | NeRF-IBVS · 12-Scenes Office5a | frequency of error reduction: 90.14% | Mixed |
| use NeRF pseudo-3D labels with IBVS refinement instead of NeRF data augmentation (DFNet) (ablation) | NeRF-IBVS · 7-Scenes average | median position/orientation error (m/deg): 0.12/3.71 -> 0.05/1.64 | Improved |
| use NeRF pseudo-3D labels with IBVS refinement instead of Direct-PoseNet (ablation) | NeRF-IBVS · 7-Scenes average | median position/orientation error (m/deg): 0.20/7.26 -> 0.05/1.64 | Improved |
| use NeRF-IBVS instead of sparse-structure augmentation FeatLoc++Au (ablation) | NeRF-IBVS · 7-Scenes average | median position/orientation error (m/deg): 0.14/5.89 -> 0.05/1.64 | Improved |
| use NeRF-IBVS instead of FeatLoc++Au on 12-Scenes (ablation) | NeRF-IBVS · 12-Scenes average | median position/orientation error (m/deg): 0.38/5.04 -> 0.02/0.95 | Improved |
| NeRF-IBVS (no GT 3D labels, ~8% of images) versus GT-3D-supervised DSAC* (ablation) | NeRF-IBVS · 7-Scenes average | median position/orientation error (m/deg): 0.05/1.64 vs 0.03/1.36 | Hurt |
| NeRF-IBVS (no GT 3D labels) versus HACNet on 12-Scenes (ablation) | NeRF-IBVS · 12-Scenes average | median position/orientation error (m/deg): 0.02/0.95 vs 0.01/0.5 | Hurt |
| NeRF-IBVS versus SA (GT 3D supervised pose regression) (ablation) | NeRF-IBVS · 7-Scenes average | median position/orientation error (m/deg): 0.15/4.55 -> 0.05/1.64 | Improved |
| use fewer posed training images (8% on 7-Scenes, 17% on 12-Scenes) (ablation) | NeRF-IBVS · 7-Scenes / 12-Scenes | number of training images: 26000 -> 2090; 16927 -> 2955 | 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.
- DFNet — DFNet: Enhance Absolute Pose Regression with Direct Feature Matching (ECCV 2022) · better on 8 of 8 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
- The Nerfect Match: Exploring NeRF Features for Visual Localization (ECCV 2024 · similarity 0.89)
- Learning Visual Servoing with Deep Features and Fitted Q-Iteration (ICLR 2017 · similarity 0.86)
- Towards Visual Localization Interoperability: Cross-Feature for Collaborative Visual Localization and Mapping (ICCV 2025 · similarity 0.86)
- IGL-Nav: Incremental 3D Gaussian Localization for Image-goal Navigation (ICCV 2025 · similarity 0.86)
- Learning Where to Look: Self-supervised Viewpoint Selection for Active Localization using Geometrical Information (ECCV 2024 · similarity 0.86)
- VTGaussian-SLAM: RGBD SLAM for Large Scale Scenes with Splatting View-Tied 3D Gaussians (ICML 2025 · 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.
