Research map / 3D vision / 3D generation
3DG-STFM: 3D Geometric Guided Student-Teacher Feature Matching
Runyu Mao, Chen Bai, Yatong An, Fengqing Zhu, Cheng Lu
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 Mutual Query Divergence distillation loss (ablation) | Unimodal LoFTR-style student · ScanNet mini (150 scenes) ablation | AUC@10deg: 31.47 -> 33.62 | Improved |
| add attentive loss on top of MQD (ablation) | student · ScanNet mini (150 scenes) ablation | AUC@10deg: 33.62 -> 34.77 | Improved |
| add MQD loss (ablation) | student · ScanNet mini (150 scenes) ablation | AUC@5deg: 14.78 -> 16.46 | Improved |
| add attentive loss on top of MQD (ablation) | student · ScanNet mini (150 scenes) ablation | AUC@20deg: 51.70 -> 52.26 | Improved |
| add depth modality in teacher (multi-modal teacher) (ablation) | unimodal model · ScanNet mini (150 scenes) ablation | AUC@10deg: 31.47 -> 36.53 | Improved |
| distill from multi-modal teacher (full student vs LoFTR) | LoFTR · ScanNet indoor pose | AUC@10deg: 40.80 -> 43.60 | Improved |
| distill from multi-modal teacher (full student vs LoFTR) | LoFTR · MegaDepth outdoor pose | AUC@10deg: 67.11 -> 68.46 | Improved |
| distill from multi-modal teacher | LoFTR · HPatches homography | AUC@5px: 71.9 -> 73.1 | Improved |
| distill from multi-modal teacher | LoFTR · HPatches homography | AUC@10px: 79.9 -> 81.0 | Improved |
| train slim model (half coarse layers) with teacher distillation (ablation) | Slim Model · ScanNet mini (150 scenes) model compression | AUC@10deg: 29.68 -> 31.76 | Improved |
| train slim model with teacher distillation vs full-size model (ablation) | Full-Size Model · ScanNet mini (150 scenes) model compression | AUC@20deg: 48.44 -> 49.51 | Improved |
| halve coarse transformer layers (4 to 2) without distillation (ablation) | Full-Size Model · ScanNet mini (150 scenes) model compression | AUC@10deg: 31.47 -> 29.68 | Hurt |
Later papers that compared with this method
1 later accepted paper reports 3DG-STFM as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- End2End Multi-View Feature Matching with Differentiable Pose Optimization (ICCV 2023 · better on 1 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
- RoMa v2: Harder Better Faster Denser Feature Matching (ECCV 2026 · similarity 0.88)
- Deep Matching Prior: Test-Time Optimization for Dense Correspondence (ICCV 2021 · similarity 0.88)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.87)
- 3DMatch: Learning Local Geometric Descriptors From RGB-D Reconstructions (CVPR 2017 · similarity 0.87)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.87)
- Self-Supervised Spatial Correspondence Across Modalities (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.
