Research map / Image generation / Image restoration
Leveraging Phase Information to Boost Unrolled Network Learning for Image Deblurring
Samira Malek, Haichuan Zhang, Chul Lee, Vishal Monga
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 UPADNet instead of EVSSM | UPADNet · GoPro | PSNR: 34.51 -> 34.63 | Improved |
| use UPADNet instead of EVSSM | UPADNet · RealBlur-J | PSNR: 34.34 -> 34.35 | Improved |
| use UPADNet instead of AdaRevD under noise | UPADNet · COCO blur+noise (sigma=0.05) | PSNR: 30.11 -> 30.86 | Improved |
| train with 60% of data instead of 100% | UPADNet · COCO | PSNR: 32.58 -> 32.12 | No clear effect |
| train with 60% of data instead of 100% | Stripformer · COCO | PSNR: 31.84 -> 27.21 | Hurt |
| train with 60% of data instead of 100% | AdaRevD · COCO | PSNR: 32.10 -> 29.51 | Hurt |
| increase UPADBlocks from 24 to 72 (ablation) | UPADNet · GoPro | PSNR: 28.41 -> 34.08 | Improved |
| increase UPADBlocks from 60 to 72 (ablation) | UPADNet · GoPro | PSNR: 33.02 -> 34.08 | Improved |
| use GELU instead of ReLU (lr 1e-3) (ablation) | UPADNet · GoPro | PSNR: 29.52 -> 34.01 | Improved |
| lower learning rate from 1e-3 to 1e-4 with GELU (ablation) | UPADNet · GoPro | PSNR: 34.01 -> 33.10 | Hurt |
| lower learning rate from 1e-3 to 1e-4 with ReLU (ablation) | UPADNet · GoPro | PSNR: 29.52 -> 31.18 | Improved |
| replace naive amplitude/phase estimators with LMMSE estimators (ablation) | LMMSE estimators · BSDS blurred test images (28) | amplitude/phase estimation error | Improved |
Closest papers by meaning
- Neural Blind Deconvolution Using Deep Priors (CVPR 2020 · similarity 0.87)
- XYDeblur: Divide and Conquer for Single Image Deblurring (CVPR 2022 · similarity 0.87)
- Deblurring Using Analysis-Synthesis Networks Pair (CVPR 2020 · similarity 0.87)
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · similarity 0.87)
- Unsupervised Blind Image Deblurring Based on Self-Enhancement (CVPR 2024 · similarity 0.86)
- Cross-Scale Self-Supervised Blind Image Deblurring via Implicit Neural Representation (NeurIPS 2024 · similarity 0.86)
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.
