Research map / Architectures and efficiency / Attention and transformers
Scalable Feature Matching via State Space Modeling and Sparse Correlation
Sin Wai Choo, Bo Li
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 |
|---|---|---|---|
| filter features by L2 norm for sparse coarse correlation versus no filtering (ablation) | SLiM · MegaDepth | inference time (ms): 128.7 -> 77.0 | Improved |
| L2 norm filtering instead of no filtering (ablation) | SLiM · MegaDepth | AUC@5: 57.9 -> 57.6 | No clear effect |
| filter by L2 norm rather than mean (ablation) | SLiM · MegaDepth | AUC@5: 55.1 -> 57.6 | Improved |
| filter by L2 norm rather than variance (ablation) | SLiM · MegaDepth | AUC@5: 57.2 -> 57.6 | Improved |
| mean-based filtering on ScanNet (ablation) | SLiM · ScanNet | AUC@5: 17.4 -> 0.6 | Hurt |
| L2 norm filtering on ScanNet (ablation) | SLiM · ScanNet | inference time (ms): 41.1 -> 23.9 | Improved |
| plug L2-norm filter into LoFTR without retraining | LoFTR · MegaDepth | AUC@5 / time: 52.8 -> 53.2; 315.4 ms -> 268.9 ms | Improved |
| plug L2-norm filter into Efficient LoFTR without retraining | Efficient LoFTR · MegaDepth | AUC@5 / time: 56.4 -> 56.1; 139.2 ms -> 76.0 ms | Mixed |
| plug L2-norm filter into MatchFormer without retraining | MatchFormer · MegaDepth | AUC@5 / time: 53.3 -> 53.4; 555.2 ms -> 487.2 ms | Improved |
| reduce recurrent refinement from 4 iterations to 1 (ablation) | SLiM · MegaDepth | AUC@5: 57.9 -> 57.6; time 77.0 -> 65.3 ms | Hurt |
| tuned RANSAC inlier threshold | SLiM · MegaDepth | AUC@5: 57.9 -> 60.1 | Improved |
| SLiM versus semi-dense methods on ScanNet accuracy | SLiM · ScanNet | AUC@5: 18.0 vs 19.6 (ASpanFormer), 19.2 (Efficient LoFTR) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- JamMa — JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba (CVPR 2025) · better on 4 of 4 shared results
- Efficient LoFTR — Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024) · better on 3 of 4 shared results
- ASpanFormer — ASpanFormer: Detector-Free Image Matching with Adaptive Span Transformer (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
- Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024 · similarity 0.88)
- JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba (CVPR 2025 · similarity 0.87)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 2024 · similarity 0.85)
- OmniGlue: Generalizable Feature Matching with Foundation Model Guidance (CVPR 2024 · similarity 0.85)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.84)
- HomoMatcher: Achieving Dense Feature Matching with Semi-Dense Efficiency by Homography Estimation (AAAI 2025 · similarity 0.84)
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.
