Research map / Image generation / Super-resolution
Iterative Soft Shrinkage Learning for Efficient Image Super-Resolution
Jiamian Wang, Huan Wang, Yulun Zhang, Yun Fu, Zhiqiang Tao
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 ISS-P instead of ASSL/SRP at pruning ratio 0.99 | SwinIR-Lightweight · Set5 x4 | PSNR: 29.15 -> 29.67 vs ASSL | Improved |
| use ISS-P at pruning ratio 0.9, x2 scale | SwinIR-Lightweight · Set5 x2 | PSNR: 37.66 (ISS-P) vs 37.69 (ASSL) | No clear effect |
| use soft percentage shrinkage (ISS-P) instead of hard thresholding (IHT) (ablation) | SwinIR-Lightweight · Set5 x4 (prune 0.9) | PSNR: 35.17 -> 35.86 | Improved |
| use ISS-P instead of ISS-R (ablation) | SwinIR-Lightweight · x4 (prune 0.9) | PSNR: 35.37 -> 35.86 | Improved |
| use ISS-R instead of IHT (ablation) | SwinIR-Lightweight · Set5 prune 0.9 | PSNR: 37.48 -> 37.38 (x2), 35.17 -> 35.37 (x4) | Mixed |
| use ISS-P on EDSR-L backbone | EDSR-L · Urban100 x4 (prune 0.95) | PSNR: 24.09 -> 24.43 vs ASSL | Improved |
| use ISS-P on CAT-R backbone | CAT-R · Urban100 x4 (prune 0.95) | PSNR: 26.09 -> 26.94 vs SRP | Improved |
| use L1-norm static pruning from random init instead of random sparse scratch training | SwinIR-Lightweight · Urban100 x4 (prune 0.9) | PSNR: 25.16 -> 25.16 | No clear effect |
| use ISS-P rather than IHT for gradient trainability (ablation) | SwinIR-Lightweight · layer-wise gradient norm and variance | gradient convergence | Improved |
Closest papers by meaning
- Learning Efficient Image Super-Resolution Networks via Structure-Regularized Pruning (ICLR 2022 · similarity 0.95)
- Aligned Structured Sparsity Learning for Efficient Image Super-Resolution (NeurIPS 2021 · similarity 0.93)
- Lightweight Image Super-Resolution via Flexible Meta Pruning (ICML 2024 · similarity 0.93)
- Global Aligned Structured Sparsity Learning for Efficient Image Super-Resolution (TPAMI 2023 · similarity 0.92)
- Structured Sparsity Learning for Efficient Video Super-Resolution (CVPR 2023 · similarity 0.90)
- Image Processing GNN: Breaking Rigidity in Super-Resolution (CVPR 2024 · similarity 0.88)
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.
