Research map / 3D vision / Depth and stereo
OANet: Learning Two-View Correspondences and Geometry Using Order-Aware Network
Jiahui Zhang, Dawei Sun, Zixin Luo, Anbang Yao, Hongkai Chen, Lei Zhou, Tianwei Shen, Yurong Chen, Long Quan, Hongen Liao
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 plain DiffUnpool layer (ablation) | PointCN baseline · YFCC100M unknown, without RANSAC | mAP@5: 23.55 -> 24.10 | No clear effect |
| add Order-Aware DiffUnpool layer (ablation) | PointCN baseline · YFCC100M unknown, without RANSAC | mAP@5: 23.55 -> 28.78 | Improved |
| replace PointCN blocks with Order-Aware Filtering blocks at second level (ablation) | PointCN + UnB · YFCC100M unknown, without RANSAC | mAP@5: 28.78 -> 32.55 | Improved |
| use larger three-level U-Net model (ablation) | PointCN + UnB + OF · YFCC100M unknown, without RANSAC | mAP@5: 32.55 -> 30.48 | Hurt |
| use geometry loss instead of L2 loss (ablation) | PointCN + UnB + OF · YFCC100M unknown, without RANSAC | mAP@5: 32.55 -> 33.68 | Improved |
| add iterative refinement network (ablation) | PointCN + UnB + OF + Geo · YFCC100M unknown, without RANSAC | mAP@5: 33.68 -> 39.33 | Improved |
| lower RANSAC inlier threshold from 0.01 to 0.001 (ablation) | PointCN baseline · YFCC100M unknown, with RANSAC | mAP@5: 17.53 -> 44.50 | Improved |
| train on 68 sequences instead of 2 (ablation) | PointCN baseline · YFCC100M unknown, with RANSAC | mAP@5: 44.50 -> 47.98 | Improved |
| use OANet instead of PointCN | OANet · YFCC100M outdoor unknown, without RANSAC | mAP@5: 23.55 -> 39.33 | Improved |
| use OANet++ instead of PointCN | OANet++ · SUN3D indoor unknown, without RANSAC | mAP@5: 9.36 -> 16.39 | Improved |
Later papers that compared with this method
4 later accepted papers report OANet as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Boosting Correspondence Learning with Structure-Aware Estimator (ECCV 2026 · better on 0 of 1 shared result)
- T-Net: Effective Permutation-Equivariant Network for Two-View Correspondence Learning (ICCV 2021 · better on 9 of 9 shared results)
- Progressive Correspondence Pruning by Consensus Learning (ICCV 2021 · better on 1 of 1 shared result)
- RANSAC-Flow: Generic Two-stage Image Alignment (ECCV 2020 · better on 1 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
- Learning Two-View Correspondences and Geometry Using Order-Aware Network (ICCV 2019 · similarity 0.97)
- T-Net: Effective Permutation-Equivariant Network for Two-View Correspondence Learning (ICCV 2021 · similarity 0.90)
- U-Match: Exploring Hierarchy-Aware Local Context for Two-View Correspondence Learning (TPAMI 2024 · similarity 0.89)
- BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024 · similarity 0.88)
- Two-View Geometry Scoring Without Correspondences (CVPR 2023 · similarity 0.87)
- Scalable and Generalizable Correspondence Pruning via Geometry-Consistent Pre-Training (TPAMI 2026 · similarity 0.87)
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.
