Research map / Image generation / Super-resolution
Self-Supervised Learning for Real-World Super-Resolution From Dual and Multiple Zoomed Observations
Zhilu Zhang, Ruohao Wang, Hongzhi Zhang, Wangmeng Zuo
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 |
|---|---|---|---|
| use the proposed self-supervised fine-tuning instead of SRA fine-tuning (ablation) | SelfTZSR++ (no alignment) · Nikon full-image | PSNR: 28.11 -> 29.14 | Improved |
| add patch-based optical flow alignment (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.19 -> 29.52 | Improved |
| add auxiliary-LR guiding alignment (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.52 -> 29.62 | Improved |
| add position preserving loss (lambda_p 0 to 100) (ablation) | auxiliary-LR generator · Nikon full-image | PSNR: 29.37 -> 29.62 | Improved |
| raise lambda_p from 100 to 10000 (ablation) | auxiliary-LR generator · Nikon full-image | PSNR: 29.62 -> 29.61 | No clear effect |
| add Gaussian noise to the auxiliary-LR (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.39 -> 29.61 | Improved |
| add only JPEG noise to the auxiliary-LR (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.39 -> 29.26 | Hurt |
| replace deformable convolution with AdaSTN (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.60 -> 29.62 | Improved |
| replace perceptual and adversarial loss with LOSW loss (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.27 -> 29.62 | Improved |
| replace SW loss with LOSW loss (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.20 -> 29.62 | Improved |
| add LOSW loss to l1 loss (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.74 -> 29.62 | Mixed |
| use both wide-angle and telephoto Refs instead of telephoto only (ablation) | SelfTZSR++ · Nikon full-image | PSNR: 29.30 -> 29.62 | Improved |
Closest papers by meaning
- Self-Supervised Learning for Real-World Super-Resolution from Dual Zoomed Observations (ECCV 2022 · similarity 0.97)
- Dual-Camera Super-Resolution With Aligned Attention Modules (ICCV 2021 · similarity 0.93)
- LMR: A Large-Scale Multi-Reference Dataset for Reference-Based Super-Resolution (ICCV 2023 · similarity 0.89)
- Reference-Based Video Super-Resolution Using Multi-Camera Video Triplets (CVPR 2022 · similarity 0.89)
- Robust Reference-Based Super-Resolution With Similarity-Aware Deformable Convolution (CVPR 2020 · similarity 0.88)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · 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.
