Research map / Image generation / GANs
Blind Image Deconvolution by Generative-based Kernel Prior and Initializer via Latent Encoding
Jiangtao Zhang, Zongsheng Yue, Hui Wang, Qian Zhao, Deyu Meng
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 pre-trained encoder kernel initializer instead of random latent initialization (ablation) | GAN kernel prior + DIP · Lai et al. benchmark | PSNR/SSIM: 19.51/0.563 -> 26.71/0.914 | Improved |
| use encoder initializer instead of fixed average-kernel initialization (ablation) | GAN kernel prior + DIP · Lai et al. benchmark | PSNR/SSIM: 23.00/0.743 -> 26.71/0.914 | Improved |
| optimize first-layer feature map w_k instead of latent code z_k (ablation) | GAN kernel prior + DIP · Lai et al. benchmark (kernel size 75) | PSNR/SSIM: 22.43/0.763 -> 26.73/0.908 | Improved |
| optimize w_k only instead of fine-tuning the whole kernel generator (ablation) | GAN kernel prior + DIP · Lai et al. benchmark (average) | PSNR/SSIM: 23.83/0.795 -> 26.71/0.914 | Improved |
| optimize w_k versus z_k on small kernels (size 31) (ablation) | GAN kernel prior + DIP · Lai et al. benchmark (kernel size 31) | PSNR/SSIM: 27.11/0.927 -> 27.89/0.932 | No clear effect |
| replace SelfDeblur random kernel init with learned generator prior and initializer | SelfDeblur · Lai et al. benchmark | convergence speed: 400 iterations vs 1000 iterations for comparable PSNR | Improved |
| replace SelfDeblur with proposed method on Lai benchmark | DIP-based BID · Lai et al. benchmark (average) | PSNR/SSIM: 21.91/0.715 -> 26.71/0.914 | Improved |
Closest papers by meaning
- Blind Image Deconvolution Using Variational Deep Image Prior (TPAMI 2023 · similarity 0.92)
- Neural Blind Deconvolution Using Deep Priors (CVPR 2020 · similarity 0.91)
- Blind Image Deconvolution by Automatic Gradient Activation (CVPR 2016 · similarity 0.90)
- Deep Learning for Handling Kernel/model Uncertainty in Image Deconvolution (CVPR 2020 · similarity 0.89)
- Kernel Diffusion: An Alternate Approach to Blind Deconvolution (ECCV 2024 · similarity 0.89)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · 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.
