Research map / 3D vision / Depth and stereo
Displacement-Invariant Cost Computation for Stereo Matching
Yiran Zhong, Charles Loop, Wonmin Byeon, Stan T. Birchfield, Yuchao Dai, Kaihao Zhang, A. Kamenev, Thomas Michael Breuel, Hongdong Li, Jan Kautz
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 |
|---|---|---|---|
| replace 3D-convolution matching on a 4D volume (Baseline3D) with shared 2D matching net per disparity | Baseline3D · SceneFlow test | runtime / EPE: 0.15 -> 0.007 s; EPE 0.78 -> 1.20 | Mixed |
| add RefineNet to the 2D matching network (ablation) | 2D matching net · SceneFlow test | EPE: 1.20 -> 1.14 | Improved |
| add RefineNet to the 2D matching network (ablation) | 2D matching net · SceneFlow test | bad-1.0: 10.31 -> 10.16 | Improved |
| feed warped right image to RefineNet (ablation) | 2D + RefineNet · SceneFlow test | EPE: 1.14 -> 1.08 | Improved |
| feed entropy map to RefineNet (ablation) | 2D + RefineNet · SceneFlow test | EPE: 1.14 -> 1.09 | Improved |
| feed entropy map to RefineNet (ablation) | 2D + RefineNet · SceneFlow test | bad-1.0: 10.16 -> 9.67 | Improved |
| use learned matching net instead of hand-crafted dot-product cost (ablation) | cost computation · SceneFlow test | EPE: 6.05 -> 1.20 | Improved |
| use 2D displacement-invariant cost instead of 3D convolution | Baseline3D · Middlebury 2014 train (cross-dataset) | bad-1.0: 53.0 -> 51.4 | Improved |
| use 2D displacement-invariant cost instead of 3D convolution | Baseline3D · KITTI 2015 train (cross-dataset) | EPE: 1.92 -> 1.63 | Improved |
| use 2D displacement-invariant cost instead of 3D convolution | Baseline3D · KITTI 2015 train (cross-dataset) | bad-1.0: 46.76% -> 37.55% | Improved |
| use 2D displacement-invariant cost instead of 3D convolution | Baseline3D · Middlebury 2014 train (cross-dataset) | avgerr: 13.5 -> 6.67 | Improved |
| displacement-invariant matching instead of context-regressing 2D net (MADNet) | MADNet · Random Dot Stereo | EPE: 51.21 -> 1.02 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- PSMnet — PSMNet: Position-Aware Stereo Merging Network for Room Layout Estimation (CVPR 2022) · better on 0 of 2 shared results
- StereoNet — StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth Prediction (ECCV 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Stereo Matching Using Multi-Level Cost Volume and Multi-Scale Feature Constancy (TPAMI 2019 · similarity 0.92)
- Displacement-Invariant Matching Cost Learning for Accurate Optical Flow Estimation (NeurIPS 2020 · similarity 0.92)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · similarity 0.91)
- Unsupervised Learning of Stereo Matching (ICCV 2017 · similarity 0.91)
- On the confidence of stereo matching in a deep-learning era: a quantitative evaluation (TPAMI 2021 · similarity 0.90)
- Learning for Disparity Estimation Through Feature Constancy (CVPR 2018 · similarity 0.90)
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.
