Research map / Architectures and efficiency / RNNs and normalization
Learning to Fuse Proposals from Multiple Scanline Optimizations in Semi-Global Matching
Johannes L. Schonberger, Sudipta N. Sinha, Marc Pollefeys
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 learned random-forest fusion of scanline proposals instead of summation | SGM · Middlebury 2014 training | bad 1px: 23.04 -> 19.99 | Improved |
| pick min-cost scanline instead of sum | SGM · Middlebury 2014 training | bad 1px: 23.04 -> 25.45 | Hurt |
| use median across scanlines instead of sum | SGM · Middlebury 2014 training | bad 1px: 23.04 -> 31.81 | Hurt |
| add the diagonal scanlines to horizontal and vertical proposals (ablation) | SGM-Forest · Middlebury 2014 training | bad 1px: 21.30 -> 20.85 | Improved |
| use all 8 scanlines instead of only top-down (ablation) | SGM-Forest · Middlebury 2014 training | bad 1px: 21.20 -> 20.85 | Improved |
| add right-view scanline proposals or filtering step (ablation) | SGM-Forest · Middlebury 2014 training | bad 4px: 4.89 -> 4.72 | Improved |
| use random forest instead of linear SVM (ablation) | SGM-SVM · Middlebury 2014 training | bad 1px: 21.88 -> 20.85 | Improved |
| use random forest instead of MLP (ablation) | SGM-MLP · Middlebury 2014 training | bad 1px: 21.83 -> 20.85 | Improved |
| train on KITTI instead of Middlebury 2005-06 (ablation) | SGM-Forest full · Middlebury 2014 training | bad 1px: 19.99 -> 20.45 | No clear effect |
| train on ETH3D instead of Middlebury 2005-06 (ablation) | SGM-Forest full · Middlebury 2014 training | bad 1px: 19.99 -> 20.32 | No clear effect |
| replace SGM with SGM-Forest using NCC cost | SGM · KITTI 2015 | bad 4px: 6.17 -> 4.17 | Improved |
| replace SGM with SGM-Forest using MC-CNN-fast | SGM · ETH3D 2017 bad 1px | bad 1px: 8.56 -> 6.08 | Improved |
Closest papers by meaning
- SGM-Nets: Semi-Global Matching With Neural Networks (CVPR 2017 · similarity 0.88)
- S2M2: Scalable Stereo Matching Model for Reliable Depth Estimation (ICCV 2025 · similarity 0.86)
- A Memory- and Accuracy-Aware Gaussian Parameter-Based Stereo Matching Using Confidence Measure (TPAMI 2019 · similarity 0.86)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.85)
- GIM: Learning Generalizable Image Matcher From Internet Videos (ICLR 2024 · similarity 0.84)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · 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.
