Research map / 3D vision / Human pose estimation
End2End Multi-View Feature Matching with Differentiable Pose Optimization
Barbara Roessle, Matthias Nießner
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 multi-view matching in the GNN (ablation) | Ours · Matterport3D 5-view | pose AUC: average drop of 14.2% when omitted | Improved |
| add end-to-end training with differentiable pose optimization (ablation) | Ours · Matterport3D 5-view | pose AUC: average drop of 7.3% when omitted | Improved |
| use weighted 8-point plus bundle adjustment instead of RANSAC at inference (ablation) | Ours w/o multi-view · ScanNet two-view | pose error AUC@5: 20.7 -> 25.7 | Improved |
| use weighted 8-point on baseline matches | LoFTR / 3DG-STFM · MegaDepth two-view | pose error AUC@5: 3DG-STFM 52.6 -> 4.0 | Hurt |
| joint multi-view matching instead of pairwise matching on 5-tuples | Ours · ScanNet runtime | matching time: 371 ms -> 338 ms (SuperGlue vs Ours) | Improved |
| use confidence thresholding before RANSAC | Ours · ScanNet runtime | RANSAC time: 126 ms -> 52 ms | Improved |
| train additionally with bundle adjustment in the loop (ablation) | Ours · ScanNet two-view | pose error AUC@5: 25.7 -> 26.0 | No clear effect |
| increase number of images in joint matching window (ablation) | Ours · ScanNet 9-image sequences | pose error AUC | Improved |
| use Ours w/o multi-view with RANSAC instead of LoFTR | LoFTR · MegaDepth two-view RANSAC | pose error AUC@5: 52.8 -> 49.5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 7 of 7 shared results
- 3DG-STFM — 3DG-STFM: 3D Geometric Guided Student-Teacher Feature Matching (ECCV 2022) · 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
- IMP: Iterative Matching and Pose Estimation With Adaptive Pooling (CVPR 2023 · similarity 0.87)
- Learning Dense Feature Matching via Lifting Single 2D Image to 3D Space (ICCV 2025 · similarity 0.86)
- CoMatcher: Multi-View Collaborative Feature Matching (CVPR 2025 · similarity 0.85)
- Learning to Filter Outlier Edges in Global SfM (CVPR 2025 · similarity 0.85)
- PoseCrafter: Extreme Pose Estimation with Hybrid Video Synthesis (NeurIPS 2025 · similarity 0.85)
- Rethinking Correspondence-based Category-Level Object Pose Estimation (CVPR 2025 · similarity 0.85)
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.
