Research map / Architectures and efficiency / CNN design
MABNet: A Lightweight Stereo Network Based on Multibranch Adjustable Bottleneck Module
Jiabin Xing, Zhi Qi, Jiying Dong, Jiaxuan Cai, Hao Liu
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 a third dilation branch (2 to 3 branches) in 2D MAB (ablation) | MABNet · SceneFlow | EPE: 1.111 -> 1.056 | Improved |
| add a second branch to a single-branch 2D MAB (ablation) | MABNet · SceneFlow | EPE: 1.072 -> 1.111 | Hurt |
| add a fourth branch (dilation 8) to 2D MAB (ablation) | MABNet · SceneFlow | EPE: 1.056 -> 1.588 | Hurt |
| add a fourth branch (dilation 8) to 2D MAB (ablation) | MABNet · KITTI2015 val | 3PE: 3.174% -> 3.225% | Hurt |
| add a third branch to a two-branch 2D MAB (ablation) | MABNet · KITTI2015 val | 3PE: 3.349% -> 3.174% | Improved |
| raise 2D scale factor lambda1 from 0.5 to 1 (ablation) | MABNet · SceneFlow | EPE: 1.072 -> 1.041 | Improved |
| raise 2D scale factor lambda3 from 0.5 to 1 (ablation) | MABNet · SceneFlow | EPE: 1.072 -> 1.161 | Hurt |
| use 3D scale factors {0.5,0.25} instead of {0.5,0.5} (ablation) | MABNet · SceneFlow | EPE: 1.006 -> 0.987 | Improved |
| use MABNet tiny instead of StereoNet | MABNet tiny · KITTI2015 benchmark | D1-all error %: 4.83 -> 3.88 | Improved |
| use MABNet tiny instead of AnyNet | MABNet tiny · KITTI2015 benchmark | D1-all error %: 6.20 -> 3.88 | Improved |
| use MABNet instead of PSMNet for runtime | MABNet · GPU runtime | runtime: 0.41s -> 0.38s | No clear effect |
| use MABNet tiny instead of StereoNet for runtime | MABNet tiny · GPU runtime | runtime: 0.015s -> 0.11s | Hurt |
Closest papers by meaning
- AANet: Adaptive Aggregation Network for Efficient Stereo Matching (CVPR 2020 · similarity 0.88)
- Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinement (CVPR 2020 · similarity 0.87)
- SMD-Nets: Stereo Mixture Density Networks (CVPR 2021 · similarity 0.87)
- GA-Net: Guided Aggregation Net for End-To-End Stereo Matching (CVPR 2019 · similarity 0.87)
- CFNet: Cascade and Fused Cost Volume for Robust Stereo Matching (CVPR 2021 · similarity 0.87)
- Generalized Binary Search Network for Highly-Efficient Multi-View Stereo (CVPR 2022 · similarity 0.87)
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.
