Research map / Image generation / Super-resolution
Uncovering the Over-Smoothing Challenge in Image Super-Resolution: Entropy-Based Quantification and Contrastive Optimization
Tianshuo Xu, Lijiang Li, Peng Mi, Xiawu Zheng, Fei Chao, Rongrong Ji, Yonghong Tian, Qiang Shen
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 DECLoss to L1 loss (ablation) | RRDB · DIV2K val | LPIPS: 0.130 -> 0.121 | Improved |
| add DECLoss to L1 loss (ablation) | RRDB · DIV2K val | PSNR: 29.47 -> 29.17 | Hurt |
| add DECLoss to L1 + perceptual loss (ablation) | RRDB · DIV2K val | LPIPS: 0.117 -> 0.108 | Improved |
| add DECLoss to L1 + perceptual loss (ablation) | RRDB · DIV2K val | PSNR: 29.10 -> 28.87 | Hurt |
| add DECLoss to ESRGAN (RaGAN) objective (ablation) | RRDB + L1 + Lp + RaGAN · DIV2K val | PSNR: 26.63 -> 28.19 | Improved |
| add DECLoss to ESRGAN (RaGAN) objective (ablation) | RRDB + L1 + Lp + RaGAN · DIV2K val | LPIPS: 0.092 -> 0.080 | Improved |
| raise positive/negative PSNR threshold from 34 to 40 (ablation) | DECLoss · DIV2K val | LPIPS: 0.1626 -> 0.1830 | Hurt |
| raise positive/negative PSNR threshold from 34 to 40 (ablation) | DECLoss · DIV2K val | PSNR: 27.860 -> 28.449 | Improved |
| vary threshold within 30-36 (ablation) | DECLoss · DIV2K val | LPIPS: 0.1634 -> 0.1630 | No clear effect |
| change batch/patch from 24/8 to 12/16 (ablation) | DECLoss · DIV2K val | LPIPS: 0.1602 -> 0.1809 | Hurt |
| change batch/patch from 24/8 to 12/16 (ablation) | DECLoss · DIV2K val | PSNR: 27.860 -> 28.230 | Improved |
| add DECLoss to Real-ESRGAN | Real-ESRGAN · DRealSR | PSNR: 25.82 -> 26.09 | Improved |
Closest papers by meaning
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.88)
- Perception-Oriented Single Image Super-Resolution Using Optimal Objective Estimation (CVPR 2023 · similarity 0.88)
- Noise-Free Optimization in Early Training Steps for Image Super-resolution (AAAI 2024 · similarity 0.87)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.87)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.87)
- Uncertainty-guided Perturbation for Image Super-Resolution Diffusion Model (CVPR 2025 · 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.
