Research map / 3D vision / Depth and stereo
GeoMoE: Divide-and-Conquer Motion Field Modeling with Mixture-of-Experts for Two-View Geometry
Jiajun Le, Jiayi Ma
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 probability injection, bi-path enhancement and F-MoE together to the baseline (ablation) | GeoMoE · YFCC100M, weighted 8-point | AUC@5deg (%): 31.69 -> 34.21 | Improved |
| add probability injection, bi-path and F-MoE (full vs baseline) (ablation) | GeoMoE · YFCC100M, weighted 8-point | AUC@20deg (%): 71.24 -> 73.53 | Improved |
| apply GeoMoE as filter to SuperPoint+NN matches | SuperPoint+NN · YFCC100M | AUC@5deg (%): 6.83 -> 29.61 | Improved |
| apply GeoMoE as filter to SuperGlue | SuperPoint+SuperGlue · YFCC100M | AUC@5deg (%): 38.24 -> 39.24 | Improved |
| apply GeoMoE as filter to LoFTR | LoFTR · YFCC100M | AUC@5deg (%): 40.42 -> 41.66 | Improved |
| replace DeMo with GeoMoE under RANSAC | relative pose · YFCC100M, RANSAC | AUC@5deg (%): 34.63 -> 34.52 | No clear effect |
| replace DeMo with GeoMoE under weighted 8-point | relative pose · YFCC100M, weighted 8-point | AUC@5deg (%): 32.57 -> 34.21 | Improved |
| replace DeMo with GeoMoE on HPatches DLT 10px | homography · HPatches DLT | Acc@10px (%): 75.69 -> 74.60 | Hurt |
| replace PointDSC with GeoMoE on 3DMatch | point cloud registration · 3DMatch | registration recall (%): 77.50 -> 78.33 | Improved |
| replace DeMo with GeoMoE (computation) | model size · YFCC100M | model size (M): 4.520 -> 5.552 | Hurt |
Closest papers by meaning
- DeMo: Deep Motion Field Consensus with Learnable Kernels for Two-view Correspondence Learning (AAAI 2025 · similarity 0.87)
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion (CVPR 2025 · similarity 0.86)
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.86)
- DeMatch: Deep Decomposition of Motion Field for Two-View Correspondence Learning (CVPR 2024 · similarity 0.86)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.86)
- Layered Motion Fusion: Lifting Motion Segmentation to 3D in Egocentric Videos (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.
