Research map / Image generation / Super-resolution
Parallax Attention for Unsupervised Stereo Correspondence Learning
Longguang Wang, Yulan Guo, Yingqian Wang, Zhengfa Liang, Zaiping Lin, Jungang Yang, Wei An
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 hourglass feature extraction instead of pyramid network (ablation) | PASMnet · SceneFlow | >3px error rate: 16.31 -> 16.09 | Improved |
| add coarse-to-fine cost regression instead of single scale (ablation) | PASMnet · SceneFlow | >3px error rate: 16.21 -> 16.09 | Improved |
| cascade 4 parallax-attention blocks instead of 1 (ablation) | PASMnet · SceneFlow | EPE: 4.68 -> 4.54 | Improved |
| add disparity refinement module (ablation) | PASMnet · SceneFlow | >3px error rate: 16.11 -> 15.91 | Improved |
| replace 4D cost volume with PAM (ablation) | PASMnet · SceneFlow | EPE: 6.02 -> 4.54 | Improved |
| replace 4D cost volume with PAM (memory) (ablation) | PASMnet · SceneFlow | memory: 1x -> 0.5x | Improved |
| add smoothness loss to photometric loss (ablation) | PASMnet · SceneFlow | EPE: 6.07 -> 4.87 | Improved |
| add PAM photometric loss (ablation) | PASMnet · SceneFlow | EPE: 4.87 -> 4.63 | Improved |
| add PAM smoothness loss (ablation) | PASMnet · SceneFlow | EPE: 4.63 -> 4.58 | Improved |
| add PAM cycle loss (ablation) | PASMnet · SceneFlow | EPE: 4.58 -> 4.54 | Improved |
| use PAM instead of cost volume at 960x540 resolution (ablation) | PASMnet · SceneFlow hard pairs | EPE: 44.04 -> 22.02 | Improved |
| use PAM instead of cost volume at 240x135 resolution (ablation) | PASMnet · SceneFlow hard pairs | >1px error rate: 39.00 -> 39.93 | Mixed |
Closest papers by meaning
- Learning Parallax Attention for Stereo Image Super-Resolution (CVPR 2019 · similarity 0.93)
- Stereo Image Restoration via Attention-Guided Correspondence Learning (TPAMI 2024 · similarity 0.93)
- Learning Stereo from Single Images (ECCV 2020 · similarity 0.88)
- Accurate and Efficient Stereo Matching via Attention Concatenation Volume (TPAMI 2023 · similarity 0.88)
- Unsupervised Learning of Stereo Matching (ICCV 2017 · similarity 0.87)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · 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.
