Research map / 3D vision / Depth and stereo
DeepPruner: Learning Efficient Stereo Matching via Differentiable PatchMatch
Shivam Duggal, Shenlong Wang, Wei-Chiu Ma, Rui Hu, Raquel Urtasun
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 confidence range predictor (ablation) | DeepPruner-Best · KITTI 2015 validation (All) | D1 outlier rate (%): 2.28 -> 1.90 | Improved |
| add confidence range predictor (ablation) | DeepPruner-Best · SceneFlow test | EPE: 0.982 -> 0.868 | Improved |
| add RefineNet (ablation) | DeepPruner-Best · KITTI 2015 validation (All) | D1 outlier rate (%): 2.32 -> 1.8 | Improved |
| add RefineNet (ablation) | DeepPruner-Best · SceneFlow test | EPE: 1.283 -> 0.858 | Improved |
| add second PatchMatch (PM-2) during SceneFlow pre-training (ablation) | DeepPruner-Best · SceneFlow test | EPE: 0.868 -> 0.858 | Improved |
| add second PatchMatch (PM-2) during SceneFlow pre-training (ablation) | DeepPruner-Best · KITTI 2015 validation (All) | D1 outlier rate (%): 1.90 -> 1.8 | Improved |
| add confidence range predictor (runtime cost) (ablation) | DeepPruner-Best · KITTI inference | runtime: 120 ms -> 172 ms | Hurt |
| downsample cost volume by 8 instead of 4 (Fast vs Best) | DeepPruner · SceneFlow test | EPE and runtime: 0.86 -> 0.97 EPE, 182 ms -> 62 ms | Mixed |
| prune disparity search space with PatchMatch and confidence range instead of full cost volume | PSM-Net style full cost volume · KITTI full-size inference | memory (MB): 4351 MB -> 1161 MB | Improved |
| remove most uncertain pixels using predicted range | DeepPruner · KITTI validation | outlier ratio: removing 6% uncertain pixels improves outlier ratio by 38% | Improved |
| DeepPruner-Fast vs prior real-time DispNetC | DeepPruner-Fast · KITTI 2015 test (All) | D1 outlier rate (%): 4.34 -> 2.59 | Improved |
Closest papers by meaning
- PatchmatchNet: Learned Multi-View Patchmatch Stereo (CVPR 2021 · similarity 0.91)
- PatchMatch-RL: Deep MVS With Pixelwise Depth, Normal, and Visibility (ICCV 2021 · similarity 0.90)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · similarity 0.89)
- Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains (CVPR 2018 · similarity 0.89)
- PCW-Net: Pyramid Combination and Warping Cost Volume for Stereo Matching (ECCV 2022 · similarity 0.89)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 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.
