Research map / 3D vision / Depth and stereo
Boosting Correspondence Learning with Structure-Aware Estimator
Tianyu Yan, Wei An, Pu Wang, Yingqian Wang
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 |
|---|---|---|---|
| replace weighted eight-point WLS with structure-aware estimator | DeMatch++ · YFCC100M relative pose | AUC@5deg: 34.33 -> 40.03 | Improved |
| replace WLS with structure-aware estimator | ConvMatch · YFCC100M relative pose | AUC@5deg: 26.78 -> 32.44 | Improved |
| replace WLS with structure-aware estimator | DeMatch · YFCC100M relative pose | inference time (ms): 10.155 -> 11.413 | Hurt |
| replace WLS with structure-aware estimator | UMatch/DeMatch/DeMatch++/ConvMatch · HPatches homography | accuracy at @1px..@10px: @3px: DeMatch++ 49.66 -> 55.34; @1px ConvMatch -1.03; @10px DeMatch -2.41 | Mixed |
| replace WLS with structure-aware estimator | DeMatch · Aachen Day-Night v1.1 night | localized queries (%): 53.4 -> 57.1 (0.25m,2deg) | Improved |
| replace WLS with structure-aware estimator | DeMatch · Aachen Day-Night v1.1 day | localized queries (%): 85.9 -> 86.0 | No clear effect |
| replace WLS with structure-aware estimator | OANet · Line fitting, outlier ratio | L2 error: 60%: 0.0254 -> 0.0260; 95%: 0.2720 -> 0.2381 | Mixed |
| replace WLS with structure-aware estimator, zero-shot indoor transfer | ConvMatch/UMatch/DeMatch/DeMo/DeMatch++ · YFCC100M to SUN3D (zero-shot) | AUC@5deg: ConvMatch SIFT 2.17 -> 2.07; UMatch 2.28 -> 2.56 | Mixed |
| fine-tune estimator on small SUN3D subset (few-shot) | structure-aware estimator · SUN3D few-shot | AUC@5deg: ConvMatch SIFT 2.49 -> 3.53 | Improved |
| add graph Laplacian operation (ablation) | DeMatch with structure-aware estimator · YFCC100M | AUC@5deg: 30.52 -> 36.58 | Improved |
| add degree matrix D (ablation) | DeMatch with structure-aware estimator · YFCC100M | AUC@5deg: 33.92 -> 36.58 | Improved |
| add learnable regularization scalar alpha (ablation) | DeMatch with structure-aware estimator · YFCC100M | AUC@5deg: 36.04 -> 36.58 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 3 on most shared results.
- ConvMatch — ConvMatch: Rethinking Network Design for Two-View Correspondence Learning (AAAI 2023) · better on 1 of 2 shared results
- BCLNet — BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024) · better on 1 of 1 shared result
- OANet — OANet: Learning Two-View Correspondences and Geometry Using Order-Aware Network (TPAMI 2020) · better on 0 of 1 shared result
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
- Scalable and Generalizable Correspondence Pruning via Geometry-Consistent Pre-Training (TPAMI 2026 · similarity 0.89)
- Progressive Correspondence Pruning by Consensus Learning (ICCV 2021 · similarity 0.87)
- Globally-Optimal Inlier Set Maximisation for Camera Pose and Correspondence Estimation (TPAMI 2018 · similarity 0.87)
- Robust Outlier Rejection for 3D Registration With Variational Bayes (CVPR 2023 · similarity 0.87)
- ECO-TR: Efficient Correspondences Finding via Coarse-to-Fine Refinement (ECCV 2022 · similarity 0.86)
- Learning Two-View Correspondences and Geometry Using Order-Aware Network (ICCV 2019 · 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.
