Research map / 3D vision / Depth and stereo
Pyramid Stereo Matching Network
Read the paperCheck 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 |
|---|---|---|---|
| Use PSMNet instead of the strongest listed baseline | Stereo disparity estimation · KITTI 2015 test (All) | D1-all (%): 2.36 -> 2.32 | Helped |
| Use PSMNet instead of the strongest listed baseline | Stereo disparity estimation · KITTI 2012 test (All, >3 px) | Error (%): 2.16 -> 1.89 | Helped |
| Use PSMNet instead of the strongest listed baseline | Stereo disparity estimation · Scene Flow test | EPE: 1.32 -> 1.09 | Helped |
| Add dilated convolution without pyramid pooling (ablation) | PSMNet basic 3D CNN · KITTI 2015 validation | 3-pixel error (%): 2.43 -> 2.16 | Helped |
| Add four-level pyramid pooling without dilated convolution (ablation) | PSMNet basic 3D CNN · KITTI 2015 validation | 3-pixel error (%): 2.43 -> 2.47 | Hurt |
| Add four-level pyramid pooling alongside dilated convolution (ablation) | PSMNet basic 3D CNN · KITTI 2015 validation | 3-pixel error (%): 2.17 -> 2.17 | No clear effect |
| Add the remaining pyramid pooling levels to 64×64 pooling with dilated convolution (ablation) | PSMNet basic 3D CNN · Scene Flow test | EPE: 1.30 -> 1.28 | Helped |
| Add stacked hourglass 3D CNN and intermediate supervision (ablation) | PSMNet with dilated convolution and four-level pyramid pooling · KITTI 2015 validation | 3-pixel error (%): 2.17 -> 1.98 | Helped |
| Use half the dilated-convolution rate (ablation) | PSMNet with stacked hourglass and four-level pyramid pooling · KITTI 2015 validation | 3-pixel error (%): 1.98 -> 1.83 | Helped |
| Use half the dilated-convolution rate (ablation) | PSMNet with stacked hourglass and four-level pyramid pooling · Scene Flow test | EPE: 1.09 -> 1.12 | Hurt |
| Set loss weights to 0.5/0.7/1.0 instead of 1.0/1.0/1.0 (ablation) | PSMNet stacked hourglass intermediate losses · KITTI 2015 validation | 3-pixel error (%): 2.01 -> 1.98 | Helped |
| Set loss weights to 0.0/0.0/1.0 instead of 0.5/0.7/1.0 (ablation) | PSMNet stacked hourglass intermediate losses · KITTI 2015 validation | 3-pixel error (%): 1.98 -> 2.49 | Hurt |
Closest papers by meaning
- Efficient Deep Learning for Stereo Matching (CVPR 2016 · similarity 0.89)
- Cost Volume Pyramid Based Depth Inference for Multi-View Stereo (CVPR 2020 · similarity 0.88)
- Practical Stereo Matching via Cascaded Recurrent Network With Adaptive Correlation (CVPR 2022 · similarity 0.87)
- StereoDRNet: Dilated Residual StereoNet (CVPR 2019 · similarity 0.87)
- Group-Wise Correlation Stereo Network (CVPR 2019 · similarity 0.87)
- IINet: Implicit Intra-inter Information Fusion for Real-Time Stereo Matching (AAAI 2024 · similarity 0.85)
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.