Research map / Image generation / Super-resolution
Unfolded Deep Kernel Estimation for Blind Image Super-Resolution
Hongyi Zheng, Hongwei Yong, Lei Zhang
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 |
|---|---|---|---|
| explicitly solve the data term in an unfolded kernel estimation network (UDKE) instead of implicit or direct DNN kernel estimation | blind SR · Set5 x2 DEPD kernels (sigma=0) | PSNR: 27.48 -> 30.50 | Improved |
| use UDKE instead of the unfolding baseline DAN | blind SR · Urban100 x2 DEPD kernels (sigma=7.65) | PSNR: 21.79 -> 24.57 | Improved |
| use UDKE kernel estimation instead of DFKP | kernel estimation · Set5 x2 (sigma=0) | Kernel PSNR: 39.4 -> 51.0 | Improved |
| use UDKE instead of DFKP on Gaussian kernels it was not designed for | blind SR · Isotropic Gaussian kernels x2 (sigma=0) | PSNR: 30.12 -> 29.80 | Hurt |
| use UDKE instead of DFKP at high noise on anisotropic Gaussian kernels | blind SR · Anisotropic Gaussian kernels x2 (sigma=7.65) | PSNR: 26.71 -> 26.73 | Improved |
| increase number of unfolding stages T up to 6 (ablation) | UDKE · Set5 x2 (sigma=0) | PSNR | Improved |
| vary trade-off parameter lambda (ablation) | UDKE · Set5 x2 (sigma=0) | PSNR | Mixed |
| use memory-efficient kernel data-term solution instead of im2col least squares | UDKE kernel solver · 2K image (2048x1024), k=11 | memory overhead: saves over 17000x memory overhead | Improved |
| replace online-training DFKP with UDKE | blind SR · Set5 x2 (sigma=0) | inference time: over 100 times faster | Improved |
| use non-blind RCAN versus blind methods (DNN and unfolding) | blind SR baselines · DEPD kernels x2 | PSNR | No clear effect |
Closest papers by meaning
- Unfolding the Alternating Optimization for Blind Super Resolution (NeurIPS 2020 · similarity 0.89)
- Deep Unfolding Network for Image Super-Resolution (CVPR 2020 · similarity 0.88)
- Deep Constrained Least Squares for Blind Image Super-Resolution (CVPR 2022 · similarity 0.88)
- Deep Plug-And-Play Super-Resolution for Arbitrary Blur Kernels (CVPR 2019 · similarity 0.88)
- Uncertainty Learning in Kernel Estimation for Multi-stage Blind Image Super-Resolution (ECCV 2022 · similarity 0.88)
- A Dynamic Kernel Prior Model for Unsupervised Blind Image 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.
