Research map / 3D vision / Depth and stereo
Learned Collaborative Stereo Refinement
Patrick Knöbelreiter, Thomas Pock
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 the RGB image data-term (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 occluded error (%): 4.43 -> 3.46 | Improved |
| add the confidence data-term (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 occluded error (%): 3.77 -> 3.46 | Improved |
| add occlusion inpainting input (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 occluded error (%): 5.12 -> 3.46 | Improved |
| add joint training (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 occluded error (%): 3.46 -> 3.37 | Improved |
| learned variational network instead of fast bilateral filter (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 occluded error (%): 7.48 -> 3.37 | Improved |
| refine WTA with variational network instead of no refinement (ablation) | VN 7,5,4 · Kitti 2015 train (ablation) | bad3 non-occluded error (%): 6.78 -> 2.55 | Improved |
| use 7 steps, 5x5 filters, 4 levels instead of 5 steps, 7x7, 3 levels (ablation) | VN · Kitti 2015 train (ablation) | bad3 non-occluded error (%): 2.58 -> 2.55 | Mixed |
| use 3x3 filters with 14 steps instead of 5x5 filters with 7 steps (ablation) | VN · Kitti 2015 train (ablation) | bad3 non-occluded error (%): 2.55 -> 3.71 | Hurt |
| use 2 levels instead of 4 levels (8 steps, 7x7) (ablation) | VN · Kitti 2015 train (ablation) | bad3 non-occluded error (%): 2.55 -> 2.97 | Hurt |
| add VN refinement on top of CNN-CRF | CNN-CRF · Kitti 2015 test | Kitti test noc error (%): 4.84 -> 4.45 | Improved |
| add VN refinement on top of CNN-CRF | CNN-CRF · Middlebury 2014 test | Middlebury test avg error: 3.02 -> 2.49 | Improved |
| add VN refinement on top of CNN-CRF | CNN-CRF · Middlebury 2014 test | Middlebury test bad4: 6.61 -> 7.71 | Hurt |
Closest papers by meaning
- Neural Disparity Refinement (TPAMI 2024 · similarity 0.89)
- Stereo Confidence Estimation via Locally Adaptive Fusion and Knowledge Distillation (TPAMI 2022 · similarity 0.88)
- Learning Stereo from Single Images (ECCV 2020 · similarity 0.88)
- Modeling Stereo-Confidence out of the End-to-End Stereo-Matching Network via Disparity Plane Sweep (AAAI 2024 · similarity 0.87)
- Adaptive Multi-Modal Cross-Entropy Loss for Stereo Matching (CVPR 2024 · 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.
