Research map / 3D vision / NeRF and radiance fields
CROSSFIRE: Camera Relocalization On Self-Supervised Features from an Implicit Representation
Arthur Moreau, Nathan Piasco, Moussab Bennehar, Dzmitry Tsishkou, Bogdan Stanciulescu, Arnaud de La Fortelle
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 |
|---|---|---|---|
| learn scene-specific self-supervised descriptors instead of memorizing pretrained extractor features (FQN-style) | CROSSFIRE · 7scenes (average) | median translation error: 0.28m / 7.3° -> 0.04m / 1.1° (vs FQN-MN) | Improved |
| use CROSSFIRE instead of FQN-MN on Cambridge Landmarks | visual relocalization · Cambridge Landmarks (average) | median translation / rotation error: 0.38m / 1.0° -> 0.37m / 1.0° | Improved |
| use CROSSFIRE instead of FQN-MN on Kings College and Shop Facade | visual relocalization · Cambridge Kings College | median translation error: 0.28m -> 0.47m (Kings College); 0.13m -> 0.20m (Shop Facade) | Hurt |
| use a constant pose prior instead of image-retrieval prior (ablation) | CROSSFIRE · 7scenes Chess | median error after 3 iterations: 0.01 / 0.4 -> 0.01 / 0.5 | No clear effect |
| add SSIM reconstruction loss (ablation) | CROSSFIRE · 7scenes Heads | median translation / rotation error: 4cm/2.1° -> 3cm/2.3° | Mixed |
| add total-variation loss (ablation) | CROSSFIRE · 7scenes Heads | median translation / rotation error: 6cm/4.0° -> 3cm/2.3° | Improved |
| use scene-specialized CROSSFIRE descriptors instead of SuperPoint for matching (ablation) | CROSSFIRE · 7scenes Chess matching accuracy | matching accuracy | Mixed |
| replace triplet loss with the proposed descriptor loss that adds a 3D-distance negative term (ablation) | CROSSFIRE · qualitative descriptor PCA | descriptor smoothness | 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 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
- Accelerated Coordinate Encoding: Learning to Relocalize in Minutes Using RGB and Poses (CVPR 2023 · similarity 0.87)
- The Nerfect Match: Exploring NeRF Features for Visual Localization (ECCV 2024 · similarity 0.86)
- BLiRF: Bandlimited Radiance Fields for Dynamic Scene Modeling (AAAI 2024 · similarity 0.84)
- Joint Optimization of Neural Radiance Fields and Continuous Camera Motion from a Monocular Video (CVPR 2025 · similarity 0.84)
- CF-NeRF: Camera Parameter Free Neural Radiance Fields with Incremental Learning (AAAI 2024 · similarity 0.84)
- Mask-Based Modeling for Neural Radiance Fields (ICLR 2024 · similarity 0.84)
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.
