Research map / Architectures and efficiency / CNN design
Efficient Deep Learning for Stereo Matching
Check 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 |
|---|---|---|---|
| Replace concatenation and subsequent comparison layers with an inner-product product layer | CNN stereo matching · KITTI 2012 validation | >3-pixel error, non-occluded (%): 12.99 -> 8.61 | Helped |
| Replace concatenation and subsequent comparison layers with an inner-product product layer | CNN stereo matching · KITTI 2012 validation | Runtime (s): 20.13 -> 0.14 | Helped |
| Train over all disparity classes with a smooth target distribution rather than independent binary match predictions | CNN stereo matching · KITTI 2012 validation | >3-pixel error, non-occluded (%): 12.99 -> 8.61 | Helped |
| Increase receptive field from 9×9 to 19×19 (ablation) | Proposed stereo matcher · KITTI 2012 validation | >3-pixel error, non-occluded (%): 16.69 -> 8.61 | Helped |
| Increase receptive field from 19×19 to 29×29 (ablation) | Proposed stereo matcher · KITTI 2012 validation | >3-pixel error, non-occluded (%): 8.61 -> 7.64 | Helped |
| Increase receptive field from 29×29 to 37×37 (ablation) | Proposed stereo matcher · KITTI 2012 validation | >3-pixel error, non-occluded (%): 7.64 -> 6.61 | Helped |
| Add cost aggregation, SGM, post-processing, and slanted-plane smoothing (ablation) | Ours(19) · KITTI 2012 validation | >3-pixel error, non-occluded (%): 8.61 -> 3.64 | Helped |
| Increase receptive field from 9×9 to 19×19 (ablation) | Proposed stereo matcher · KITTI 2015 validation | >3-pixel error, non-occluded (%): 15.25 -> 8.95 | Helped |
| Increase receptive field from 19×19 to 29×29 (ablation) | Proposed stereo matcher · KITTI 2015 validation | >3-pixel error, non-occluded (%): 8.95 -> 7.23 | Helped |
| Increase receptive field from 29×29 to 37×37 (ablation) | Proposed stereo matcher · KITTI 2015 validation | >3-pixel error, non-occluded (%): 7.23 -> 7.13 | Helped |
| Add cost aggregation, SGM, post-processing, and slanted-plane smoothing (ablation) | Ours(19) · KITTI 2015 validation | >3-pixel error, non-occluded (%): 8.95 -> 4.20 | Helped |
| Add SGM to the CNN unary costs (ablation) | Ours(19) · KITTI 2012 validation | >3-pixel error, non-occluded (%): 8.61 -> 7.48 | Helped |
Closest papers by meaning
- Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks (AAAI 2022 · similarity 0.91)
- Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains (CVPR 2018 · similarity 0.91)
- Learning for Disparity Estimation Through Feature Constancy (CVPR 2018 · similarity 0.90)
- Practical Stereo Matching via Cascaded Recurrent Network With Adaptive Correlation (CVPR 2022 · similarity 0.90)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.89)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · similarity 0.88)
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.