Research map / 3D vision / Depth and stereo
SC-Net: Robust Correspondence Learning via Spatial and Cross-Channel Context
Shuyuan Lin, Hailiang Liao, Qiang Qi, Junjie Huang, Taotao Lai, Jian Weng
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 hierarchical encoder-decoder (HED) to the ConvMatch baseline (ablation) | SC-Net baseline · YFCC100M known scenes | mAP@5deg: 46.09 -> 57.96 | Improved |
| add the motion feature modulator (MFM) on top of HED (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 57.96 -> 59.60 | Improved |
| add soft filtering with previous-layer classification logits (AFR) (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 59.60 -> 61.96 | Improved |
| add position-aware attention (AFR) (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 61.96 -> 64.35 | Improved |
| use 6 rectifying layers instead of 4 (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 60.79 -> 64.35 | Improved |
| increase rectifying layers from 6 to 8 (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 64.35 -> 63.96 | Hurt |
| use grid size K=16 instead of K=12 (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 63.24 -> 64.35 | Improved |
| increase grid size from 16 to 20 (ablation) | SC-Net · YFCC100M known scenes | mAP@5deg: 64.35 -> 63.40 | Hurt |
| train on SIFT, test on SuperPoint matches with SC-Net instead of ConvMatch+ | correspondence pruning · YFCC100M SuperPoint (trained on SIFT) | mAP@5deg: 30.85 -> 44.08 | Improved |
| SC-Net instead of best prior on outlier removal | correspondence pruning · YFCC100M known scenes | F-score: 79.87 -> 82.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BCLNet — BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024) · better on 8 of 8 shared results
- ConvMatch+ — ConvMatch: Rethinking Network Design for Two-View Correspondence Learning (AAAI 2023) · better on 2 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
- DeMo: Deep Motion Field Consensus with Learnable Kernels for Two-view Correspondence Learning (AAAI 2025 · similarity 0.87)
- SCNet: Learning Semantic Correspondence (ICCV 2017 · similarity 0.87)
- BCLNet: Bilateral Consensus Learning for Two-View Correspondence Pruning (AAAI 2024 · similarity 0.87)
- DeMatch: Deep Decomposition of Motion Field for Two-View Correspondence Learning (CVPR 2024 · similarity 0.86)
- SCFlow2: Plug-and-Play Object Pose Refiner with Shape-Constraint Scene Flow (CVPR 2025 · similarity 0.85)
- Learning Structure-from-Motion with Graph Attention Networks (CVPR 2024 · 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.
