Research map / Image generation / Super-resolution
"Restore Globally, Refine Locally: A Mask-Guided Scheme to Accelerate Super-Resolution Networks"
Xiaotao Hu, Jun Xu, Shuhang Gu, Ming-Ming Cheng, Li Liu
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 two-channel softmax spatial feature in the MP module (ablation) | PAN-M · Urban100 (x4) | PSNR: 25.96 -> 26.02 | Improved |
| use the three-step training strategy instead of end-to-end (ablation) | FSRCNN-M · Urban100 (x2) | PSNR: 29.10 -> 29.56 | Improved |
| add supervision of the MP module with the coarse-final difference (ablation) | FSRCNN-M · Urban100 (x2) | PSNR: 29.50 -> 29.56 | Improved |
| select largest errors rather than random (ablation) | RFDN-M · Urban100 (x2) | PSNR: 32.14 -> 32.36 | Improved |
| select largest errors rather than smaller (ablation) | RFDN-M · Urban100 (x2) | PSNR: 32.03 -> 32.36 | Improved |
| refine more patches (K=All/2 instead of K=1000, largest errors) (ablation) | RFDN-M · Urban100 (x2) | PSNR: 32.26 -> 32.36 | Improved |
| give Refine-Net 2 blocks instead of 6 (14+2 vs 10+6) (ablation) | PAN-M · Urban100 (x2) | PSNR: 31.67 -> 31.90 | Improved |
| increase feature patch size p from 4 to 8 (ablation) | FSRCNN-M · Urban100 (x4) | PSNR: 24.50 -> 24.48 | Hurt |
| replace ClassSR with MGA acceleration | RCAN · Urban100 (x4, RGB) | PSNR: 25.14 -> 25.43 | Improved |
| replace ClassSR with MGA acceleration | RCAN · Kirin 980 mobile | inference time: 38.91s -> 35.51s | Improved |
| apply MGA to decomposed network | RCAN-D · Urban100 (x4) | FLOPs (G): 808.21 -> 728.52 | Improved |
| apply MGA to decomposed network | PAN-D · Urban100 (x4) | FLOPs (G): 36.05 -> 29.74 | Improved |
Closest papers by meaning
- Accelerating the Super-Resolution Convolutional Neural Network (ECCV 2016 · similarity 0.92)
- ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic (CVPR 2021 · similarity 0.89)
- FSR: A General Frequency-Oriented Framework to Accelerate Image Super-resolution Networks (AAAI 2023 · similarity 0.89)
- Image Super-Resolution Using Deep Convolutional Networks (TPAMI 2016 · similarity 0.89)
- Image Processing GNN: Breaking Rigidity in Super-Resolution (CVPR 2024 · similarity 0.88)
- Efficient Look-Up Table from Expanded Convolutional Network for Accelerating Image Super-resolution (AAAI 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.
