Research map / 3D vision / NeRF and radiance fields
The Nerfect Match: Exploring NeRF Features for Visual Localization
Qunjie Zhou, Maxim Maximov, Or Litany, Laura Leal-Taixé
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 |
|---|---|---|---|
| use NeRF inner features (layers 1-6) instead of positional-encoded 3D points for 2D-3D matching (ablation) | NeRFMatch-Mini · Cambridge Landmarks | localization recall (%): 51.4 -> 59.2 | Improved |
| use the last NeRF layer feature f7 instead of middle layers (ablation) | NeRFMatch-Mini · Cambridge Landmarks | median translation (cm): 27.9 -> 61.3 | Hurt |
| match against raw 3D coordinates instead of positional encoding (ablation) | NeRFMatch-Mini · Cambridge Landmarks | median translation (cm): 34.3 -> 458.0 | Hurt |
| use ImageNet-pretrained ConvFormer backbone instead of from scratch (ablation) | NeRFMatch-Mini · Cambridge Landmarks | avg recall (%): 50.7 -> 59.2 | Improved |
| use coarse-to-fine attention matcher instead of minimal matcher (ablation) | NeRFMatch · Cambridge Landmarks | avg recall (%): 59.2 -> 71.3 | Improved |
| train on multiple scenes instead of per-scene (ablation) | NeRFMatch · Cambridge Landmarks | avg recall (%): 71.3 -> 65.2 | Hurt |
| use iterative refinement instead of photometric optimization refinement (ablation) | NeRFMatch (full) · Cambridge Landmarks | latency per step (ms): 398.4 ms -> 141.2 ms | Improved |
| use Instant NGP instead of MipNeRF as NeRF (ablation) | NeRFMatch · Cambridge Landmarks | avg median error (cm/deg)/recall: 13.3/0.3/80.8 -> 28.1/0.5/61.3 | Hurt |
| retrieve with NeRF-rendered images instead of real database images (ablation) | NeRFMatch · Cambridge Landmarks | avg median error (cm): 13.3 -> 14.6 | Hurt |
| apply NeRF features for matching on indoor 7-Scenes | NeRFMatch · 7-Scenes | avg recall (%): 97.8 -> 78.2 | Hurt |
Closest papers by meaning
- Nerflets: Local Radiance Fields for Efficient Structure-Aware 3D Scene Representation From 2D Supervision (CVPR 2023 · similarity 0.91)
- PNeRFLoc: Visual Localization with Point-Based Neural Radiance Fields (AAAI 2024 · similarity 0.90)
- D-NeRF: Neural Radiance Fields for Dynamic Scenes (CVPR 2021 · similarity 0.90)
- To View Transform or Not to View Transform: NeRF-based Pre-training Perspective (ICLR 2026 · similarity 0.90)
- AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware Training (CVPR 2023 · similarity 0.90)
- NeRF Analogies: Example-Based Visual Attribute Transfer for NeRFs (CVPR 2024 · 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.
