Research map / Image generation / Super-resolution
Rethinking the Upsampling Process in Light Field Super-Resolution with Spatial-Epipolar Implicit Image Function
Ruixuan Cong, Yu Wang, Mingyuan Zhao, Da Yang, Rongshan Chen, Hao Sheng
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 original upsampling with SEIIF | LF-InterNet · LFSSR average (x4, fixed-scale) | PSNR (dB): 31.61 -> 32.12 | Improved |
| replace original upsampling with SEIIF | LFT · LFSSR average (x2, fixed-scale) | PSNR (dB): 38.85 -> 38.97 | Improved |
| replace original upsampling with SEIIF across eight encoders | eight LFSSR encoders · LFSSR average (x2) | PSNR (dB): [0.12, 0.39] | Improved |
| replace original upsampling with SEIIF across eight encoders | eight LFSSR encoders · LFSSR average (x4) | PSNR (dB): [0.23, 0.51] | Improved |
| add SIIF branch to SEIIF (ablation) | SEIIF · SEIIF ablation (DistgSSR, x2) | PSNR (dB): 38.92 -> 39.01 | Improved |
| add EIIF branch to SEIIF (ablation) | SEIIF · SEIIF ablation (DistgSSR, x2) | PSNR (dB): 38.87 -> 39.01 | Improved |
| add cross-branch feature interaction with two-step upsampling (ablation) | SEIIF · SEIIF ablation (DistgSSR, x2) | PSNR (dB): 38.96 -> 39.01 | Improved |
| use oriented line sampling instead of nearest sampling in EIIF (ablation) | SEIIF · EIIF sampling ablation (x2) | PSNR (dB): 38.95 -> 39.01 | Improved |
| use local sampling instead of nearest sampling in EIIF (ablation) | SEIIF · EIIF sampling ablation (x2) | PSNR (dB): 38.95 -> 38.94 | Hurt |
| use attention weighting instead of variance in oriented line sampling (ablation) | SEIIF · EIIF sampling ablation (x4) | PSNR (dB) and FLOPs: 32.53 -> 32.53; FLOPs 27.3 -> 60.3 | No clear effect |
| train with discrete scales U(1,4,0.5) instead of continuous U(1,4) (ablation) | SEIIF arbitrary-scale · Training strategy ablation (x2) | PSNR (dB): 38.35 -> 39.01 | Improved |
| add fine-tuning on target scale after multi-scale training (ablation) | SEIIF fixed-scale · Training strategy ablation (x2) | PSNR (dB): 38.95 -> 39.15 | Improved |
Closest papers by meaning
- Learning Three-Domain Implicit Image Function for Arbitrary-Scale Light Field Super-Resolution (TPAMI 2026 · similarity 0.97)
- Learning Non-Local Spatial-Angular Correlation for Light Field Image Super-Resolution (ICCV 2023 · similarity 0.91)
- Light Field Reconstruction Using Convolutional Network on EPI and Extended Applications (TPAMI 2018 · similarity 0.88)
- Diving Into Epipolar Transformers for Light Field Super-Resolution and Disparity Estimation (TPAMI 2026 · similarity 0.87)
- Occlusion-Embedded Hybrid Transformer for Light Field Super-Resolution (AAAI 2025 · similarity 0.85)
- Fast Omni-Directional Image Super-Resolution: Adapting the Implicit Image Function with Pixel and Semantic-Wise Spherical Geometric Priors (AAAI 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.
