Research map / Image generation / Super-resolution
Reflash Dropout in Image Super-Resolution
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 channel-wise dropout before the last convolution (p=0.7) to Real-SRResNet | Real-SRResNet · Set5 clean ×4 | PSNR: 24.89 -> 25.67 (+0.78) | Helped |
| Add channel-wise dropout before the last convolution (p=0.5) to Real-RRDB | Real-RRDB · Set5 blur ×4 | PSNR: 25.14 -> 26.09 (+0.95) | Helped |
| Add channel-wise dropout before the last convolution | Real-SRResNet and Real-RRDB · NTIRE 2018 realistic mild/difficult/wild data | average PSNR: +0.16, +0.20, +0.17 for R-SRResNet; +0.52, +0.50, +0.53 for R-RRDB | Helped |
| Add channel-wise dropout before the last convolution (p=0.5) | Real-SwinIR · Multi-degradation SR benchmarks | PSNR: 0.46 maximum improvement; -0.10 to +0.46 across reported cells | Mixed |
| Use channel-wise dropout before the last convolution rather than no dropout (ablation) | SRResNet · Single-degradation bicubic SR | PSNR | No clear effect |
| Apply channel-wise dropout after every convolution instead of only before the last convolution (ablation) | SRResNet · Single-degradation bicubic SR | PSNR | Hurt |
| Use element-wise dropout instead of channel-wise dropout (ablation) | SR networks · Single-degradation bicubic SR | PSNR | Hurt |
| Use dropout after B12 rather than before the last convolution (ablation) | Real-SRResNet · Multi-degradation ×4 SR | PSNR | Mixed |
| Replace only the last four residual blocks with dropped residual blocks (ablation) | Real-SRResNet · Multi-degradation ×4 SR | PSNR | Hurt |
| Replace the second half of residual blocks with dropped residual blocks (ablation) | Real-SRResNet · Multi-degradation ×4 SR | PSNR | Hurt |
| Replace all residual blocks with dropped residual blocks (ablation) | Real-SRResNet · Multi-degradation ×4 SR | PSNR | Hurt |
| Increase dropout probability from p=0.1 to p=0.9 (ablation) | Real-SRResNet · Manga109 noise and Urban100 noise | PSNR: 22.31 vs 22.57; 20.64 vs 20.97 | Mixed |
Closest papers by meaning
- Navigating Beyond Dropout: An Intriguing Solution towards Generalizable Image Super Resolution (CVPR 2024 · similarity 0.91)
- Adaptive Dropout: Unleashing Dropout across Layers for Generalizable Image Super-Resolution (CVPR 2025 · similarity 0.89)
- Image Processing GNN: Breaking Rigidity in Super-Resolution (CVPR 2024 · similarity 0.85)
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.85)
- DORNet: A Degradation Oriented and Regularized Network for Blind Depth Super-Resolution (CVPR 2025 · similarity 0.85)
- FIPER: Factorized Features for Robust Image Super-Resolution and Compression (NeurIPS 2025 · similarity 0.84)
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.