Research map / Image generation / Image restoration
Rethinking Deep Image Prior for Denoising
Yeonsik Jo, Se Young Chun, Jonghyun Choi
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 stochastic temporal ensembling (noise regularization with random level plus EMA) (ablation) | SURE-based DIP · CSet9 (sigma=25) | PSNR: 31.34 -> 31.54 | Improved |
| add stochastic temporal ensembling (ablation) | SURE-based DIP · CSet9 (sigma=25) | SSIM: 0.955 -> 0.953 | Hurt |
| add stochastic temporal ensembling (ablation) | SURE-based DIP · CSet9 (sigma=25) | LPIPS: 0.108 -> 0.107 | No clear effect |
| use hyperparameter-free MC divergence instead of epsilon-dependent approximation (DIP-SURE*) | DIP with SURE · CSet9 (sigma=25) | PSNR: 30.33 -> 31.34 | Improved |
| replace DIP early stopping with SURE-based objective and zero-crossing stop | DIP · CSet9 (sigma=25) | PSNR: 29.96 -> 31.54 | Improved |
| replace DIP early stopping with SURE-based objective and zero-crossing stop | DIP · CSet9 (sigma=25) | LPIPS: 0.152 -> 0.107 | Improved |
| use proposed method instead of Self2Self at low noise | Self2Self · CSet9 (sigma=15) | LPIPS: 0.135 -> 0.070 | Improved |
| use proposed method instead of Self2Self at high noise | Self2Self · CSet9 (sigma=50) | PSNR: 29.25 -> 28.90 | Hurt |
| increase number of Self2Self ensemble members | Self2Self · CSet9 | PSNR vs LPIPS | Mixed |
| use proposed Poisson extension instead of DIP at high noise | DIP · Poisson (zeta=0.2) | PSNR: 21.43 -> 22.85 | Improved |
| use proposed Poisson extension instead of Self2Self at low noise | Self2Self · Poisson (zeta=0.01) | PSNR: 32.18 -> 32.00 | Hurt |
| use proposed method instead of Self2Self for runtime | Self2Self · CSet9 | inference time: 35 hours -> 4 hours | Improved |
Closest papers by meaning
- A Principled Self-Referenced Early Stopping Approach for Deep Image Prior (NeurIPS 2026 · similarity 0.88)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.88)
- Stochastic Deep Restoration Priors for Imaging Inverse Problems (ICML 2025 · similarity 0.88)
- Exact Posterior Score Estimation for Solving Linear Inverse Problems (NeurIPS 2026 · similarity 0.88)
- Image Reconstruction Via Autoencoding Sequential Deep Image Prior (NeurIPS 2024 · similarity 0.88)
- Realistic Noise Synthesis with Diffusion Models (AAAI 2025 · 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.
