Research map / 3D vision / Depth and stereo
A Benchmark for Heterogeneous Stereo Deblurring with Physically- and Epipolar-constrained Cross Attention
Jiah Kim, Hoju Shin, Seung-Wook Kim, Seowon Ji
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 PECA cross-view attention | XYDeblur · HSD test | PSNR: 30.76 -> 32.22 | Improved |
| add PECA cross-view attention | Restormer · HSD test | PSNR: 30.94 -> 32.47 | Improved |
| add PECA cross-view attention | NAFNet · HSD test | PSNR: 32.16 -> 32.92 | Improved |
| use PECA instead of naive stereo concatenation (ablation) | NAFNet · HSD test | PSNR: 32.24 -> 32.92 | Improved |
| use PECA instead of naive stereo concatenation (ablation) | Restormer · HSD test | PSNR: 31.83 -> 32.47 | Improved |
| scale single-view capacity (channel expansion) instead of adding the reference view (ablation) | XYDeblur/Restormer/NAFNet · HSD test | PSNR: 30.76 -> 30.92 (XYDeblur); 32.16 -> 32.28 (NAFNet) | Improved |
| add the sharp wide-view reference by naive stereo concatenation (ablation) | XYDeblur · HSD test | PSNR: 30.76 -> 31.96 | Improved |
| replace global 2D cross attention with PECA bounded window (ablation) | XYDeblur · HSD test | PSNR: 30.72 -> 32.22 | Improved |
| restrict attention to full scanline instead of global 2D (ablation) | XYDeblur · HSD test | PSNR: 30.72 -> 30.75 | No clear effect |
| restrict attention to bounded epipolar window instead of full row (ablation) | XYDeblur · HSD test | attention module MACs (G): 7.6 -> 2.9 | Improved |
| add PECA instead of naive concatenation under heavy occlusion (Top 10% holes) (ablation) | XYDeblur · HSD test (Top 10% hole ratio) | PSNR: 28.03 -> 28.35 (PECA); 28.03 -> 27.93 (concat) | Improved |
| use PECA instead of naive concatenation on hardest samples (ablation) | NAFNet · HSD test (Hard subset) | PSNR: 27.23 -> 28.99 | Improved |
Closest papers by meaning
- DAVANet: Stereo Deblurring With View Aggregation (CVPR 2019 · similarity 0.88)
- Asymmetric Dual-Lens Video Deblurring (NeurIPS 2025 · similarity 0.87)
- Simultaneous Stereo Video Deblurring and Scene Flow Estimation (CVPR 2017 · similarity 0.87)
- Stereo Anywhere: Robust Zero-Shot Deep Stereo Matching Even Where Either Stereo or Mono Fail (CVPR 2025 · similarity 0.86)
- Mono2Stereo: A Benchmark and Empirical Study for Stereo Conversion (CVPR 2025 · similarity 0.85)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · 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.
