Research map / Label-efficient and robust learning / Recognition and long-tail
Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 full multi-scale, multi-level LSP and CSR | GwcNet · SceneFlow test | EPE (px): 1.04 -> 0.75 | Helped |
| Add full multi-scale, multi-level LSP and CSR | GwcNet · KITTI 2015 validation | D1-all (%): 1.65 -> 1.30 | Helped |
| Add LSP to convolutional features (ablation) | GwcNet · SceneFlow test | EPE (px): 1.04 -> 0.95 | Helped |
| Use single-scale rather than full multi-scale, multi-level LSP (ablation) | GwcNet · SceneFlow test | EPE (px): 0.95 -> 0.97 | Hurt |
| Use single-level rather than full multi-scale, multi-level LSP (ablation) | GwcNet · SceneFlow test | EPE (px): 0.95 -> 0.99 | Hurt |
| Replace no refinement with DSR (ablation) | GwcNet · SceneFlow test | EPE (px): 1.04 -> 0.80 | Helped |
| Replace no refinement with CSR (ablation) | GwcNet · SceneFlow test | EPE (px): 1.04 -> 0.79 | Helped |
| Use ConvNet disparity-residual refinement instead of no refinement (ablation) | GwcNet · SceneFlow test | >1px (%): 7.68 -> 8.05 | Hurt |
| Increase CSR assembled point number from N=0 to N=8 (ablation) | GwcNet · SceneFlow | EPE (px): 0.95 -> 0.72 | Helped |
| Increase CSR assembled point number from N=0 to N=8 (ablation) | GwcNet · SceneFlow | Memory (G): 3.3 -> 9.7 | Hurt |
| Add LSP (ablation) | GwcNet · Middlebury | Bad2.0-noc (%): 22.2 -> 20.1 | Helped |
| Add CSR after LSP (ablation) | GwcNet · ETH3D | Bad1.0-noc (%): 9.6 -> 9.2 | Helped |
Closest papers by meaning
- Learning for Disparity Estimation Through Feature Constancy (CVPR 2018 · similarity 0.91)
- Practical Stereo Matching via Cascaded Recurrent Network With Adaptive Correlation (CVPR 2022 · similarity 0.91)
- Efficient Deep Learning for Stereo Matching (CVPR 2016 · similarity 0.91)
- Deep Stereo Matching With Explicit Cost Aggregation Sub-Architecture (AAAI 2018 · similarity 0.90)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.88)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · 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.