Research map / 3D vision / Human pose estimation
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
Shuai Chen, Xinghui Li, Zirui Wang, Victor Adrian Prisacariu
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 Random View Synthesis (online NeRF-rendered training views) (ablation) | DFNet · Shop Facade | translation/rotation error: 1.49m/5.80 -> 0.86m/4.05 | Improved |
| replace original triplet loss with customized in-triplet mining loss (ablation) | DFNet · Shop Facade | translation/rotation error: 0.86m/4.05 -> 0.72m/2.58 | Improved |
| add direct feature matching using NeRF-W renderer (ablation) | DFNet · Shop Facade | translation/rotation error: 0.72m/2.58 -> 0.43m/1.62 | Improved |
| use histogram-assisted NeRF instead of NeRF-W for direct feature matching (ablation) | DFNet_dm · Shop Facade | translation/rotation error: 0.43m/1.62 -> 0.15m/0.65 | Improved |
| add histogram-assisted NeRF to photometric direct matching (ablation) | Direct-PN+U · Shop Facade | translation/rotation error: 1.41m/6.97 -> 0.72m/3.39 | Improved |
| add middle-level features to direct matching (ablation) | DFNet_dm · Shop Facade | translation error: 0.15m -> 0.19m | Hurt |
| add middle and coarse features to direct matching (ablation) | DFNet_dm · Shop Facade | translation error: 0.15m -> 0.20m | Hurt |
| replace triplet loss with MSE loss (ablation) | DFNet feature extractor · Kings (Cambridge) | feature quality | Hurt |
| add unlabeled feature-metric direct matching finetuning (ablation) | DFNet · 7-Scenes (avg) | translation/rotation error: 0.12/3.71 -> 0.07/2.21 | Improved |
| add unlabeled feature-metric direct matching finetuning (ablation) | DFNet · Cambridge Landmarks (avg) | translation/rotation error: 1.19/2.90 -> 0.39/0.96 | Improved |
| swap VGG16 backbone for EfficientNet-B0 (ablation) | DFNet · 7-Scenes (avg) | translation/rotation error: 0.12/3.71 -> 0.08/3.47 | Improved |
| use online RVS instead of offline LENS-style generation (EfficientNet-B0 vs EB3) | DFNet · 7-Scenes (avg) | translation/rotation error: 0.08/3.47 vs 0.08/3.00 | No clear effect |
Later papers that compared with this method
6 later accepted papers report DFNet as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Rethinking Pose Refinement in 3D Gaussian Splatting under Pose Prior and Geometric Uncertainty (CVPR 2026 · better on 1 of 1 shared result)
- Adversarial Exploitation of Data Diversity Improves Visual Localization (ICCV 2025 · better on 2 of 2 shared results)
- GS-CPR: Efficient Camera Pose Refinement via 3D Gaussian Splatting (ICLR 2025 · better on 1 of 1 shared result)
- Learning Neural Volumetric Pose Features for Camera Localization (ECCV 2024 · better on 3 of 3 shared results)
- NeRF-IBVS: Visual Servo Based on NeRF for Visual Localization and Navigation (NeurIPS 2023 · better on 8 of 8 shared results)
- CROSSFIRE: Camera Relocalization On Self-Supervised Features from an Implicit Representation (ICCV 2023 · better on 2 of 2 shared results)
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.
- MapNet — MapNet: An Allocentric Spatial Memory for Mapping Environments (CVPR 2018) · 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
- Map-Relative Pose Regression for Visual Re-Localization (CVPR 2024 · similarity 0.89)
- Neural Refinement for Absolute Pose Regression with Feature Synthesis (CVPR 2024 · similarity 0.88)
- ConDo: Continual Domain Expansion for Absolute Pose Regression (AAAI 2025 · similarity 0.88)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.88)
- Scene-agnostic Pose Regression for Visual Localization (CVPR 2025 · similarity 0.86)
- FAR: Flexible Accurate and Robust 6DoF Relative Camera Pose Estimation (CVPR 2024 · similarity 0.86)
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.
