Research map / 3D vision / Human pose estimation
SemLight: Distilled Semantic–Geometric Fusion for Efficient Local Feature Matching
Guanglu Shi, Xiangzeng Liu, Yunan LI, Tuo Pang, Qiguang Miao
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 DistilSem with naive addition to LiftFeat (ablation) | LiftFeat · MegaDepth-1500 | AUC@5: 44.7 -> 47.8 | Improved |
| add reliability-gated fusion on top of DistilSem (ablation) | LiftFeat+DistilSem · MegaDepth-1500 | AUC@5: 47.8 -> 49.4 | Improved |
| add cross-modal token attention on top of DistilSem (ablation) | LiftFeat+DistilSem · MegaDepth-1500 | AUC@5: 47.8 -> 48.9 | Improved |
| combine gating and cross-modal attention (RGFusion) (ablation) | LiftFeat+DistilSem · MegaDepth-1500 | AUC@5: 49.4 -> 50.2 | Improved |
| combine gating and cross-modal attention (RGFusion) (ablation) | LiftFeat+DistilSem · MegaDepth-1500 | AUC@5: 48.9 -> 50.2 | Improved |
| use frozen DINOv3 instead of distilled semantic head (ablation) | SemLight · MegaDepth-1500 | AUC@5: 50.2 -> 53.6 | Improved |
| use frozen DINOv3 instead of distilled semantic head (ablation) | SemLight · MegaDepth-1500 | GPU time (ms): 5.4 -> 738.8 | Hurt |
| full SemLight versus LiftFeat | SemLight · MegaDepth-1500 | AUC@5: 44.7 -> 50.2 | Improved |
| full SemLight versus XFeat | SemLight · MegaDepth-1500 | AUC@5: 42.6 -> 50.2 | Improved |
| SemLight versus Cadar et al. (trained on ScanNet) | SemLight · ScanNet-1500 | AUC@5: 20.3 -> 19.2 | Hurt |
| SemLight versus LiftFeat on viewpoint homography | SemLight · HPatches viewpoint | MHA@3: 71.1 -> 72.1 | Improved |
| SemLight versus LiftFeat on viewpoint homography | SemLight · HPatches viewpoint | MHA@7: 87.5 -> 87.2 | No clear effect |
Closest papers by meaning
- Matching While Perceiving: Enhance Image Feature Matching with Applicable Semantic Amalgamation (AAAI 2025 · similarity 0.88)
- TextFM: Robust Semi-dense Feature Matching with Language Guidance (CVPR 2026 · similarity 0.88)
- GeoMix: Descriptor-Free Visual Localization via Global Context and Multi-Detector Training (ECCV 2026 · similarity 0.88)
- SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation (CVPR 2024 · similarity 0.87)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.86)
- Bridging Viewpoint Gaps: Geometric Reasoning Boosts Semantic Correspondence (CVPR 2025 · 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.
