Research map / Architectures and efficiency / CNN design
End-To-End Training of Hybrid CNN-CRF Models for Stereo
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 |
|---|---|---|---|
| Add CRF inference with fixed contrast-sensitive pairwise weights (ablation) | Unary-CNN · Middlebury V3 design set | bad4 (%): 23.89 -> 11.18 | Helped |
| Add CRF inference with fixed contrast-sensitive pairwise weights (ablation) | Unary-CNN · Kitti 2015 design set, non-occluded pixels | bad3 (%): 28.38 -> 6.33 | Helped |
| Jointly train Unary-CNN and CRF parameters (ablation) | CNN3+CRF · Middlebury V3 design set | bad4 (%): 11.18 -> 9.48 | Helped |
| Jointly train Unary-CNN and CRF parameters (ablation) | CNN7+CRF · Kitti 2015 design set, all pixels | bad3 (%): 5.85 -> 5.98 | Hurt |
| Add jointly trained Pairwise-CNN edge weights (ablation) | CNN3 Joint · Kitti 2015 design set, non-occluded pixels | bad3 (%): 6.11 -> 4.75 | Helped |
| Add jointly trained Pairwise-CNN edge weights (ablation) | CNN3 Joint · Middlebury V3 design set | bad4 (%): 9.48 -> 9.45 | No clear effect |
| Use 7 rather than 3 Unary-CNN layers (ablation) | Pixel-wise Unary-CNN · Middlebury V3 design set | bad4 (%): 23.89 -> 18.58 | Helped |
| Use 7 rather than 3 Unary-CNN layers in the full model (ablation) | CNNx PW+Joint · Kitti 2015 design set, non-occluded pixels | bad3 (%): 4.75 -> 4.04 | Helped |
| Use a model trained on Kitti rather than Middlebury (ablation) | Hybrid CNN-CRF · Middlebury V3 design set | bad4 (%): 7.88 -> 15.22 | Hurt |
Closest papers by meaning
- Multi-Scale Continuous CRFs as Sequential Deep Networks for Monocular Depth Estimation (CVPR 2017 · similarity 0.88)
- Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains (CVPR 2018 · similarity 0.88)
- Structured Attention Guided Convolutional Neural Fields for Monocular Depth Estimation (CVPR 2018 · similarity 0.87)
- Neural Markov Random Field for Stereo Matching (CVPR 2024 · similarity 0.87)
- Efficient Deep Learning for Stereo Matching (CVPR 2016 · similarity 0.87)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.86)
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.