Research map / Image generation / Image restoration
Iterative Denoiser and Noise Estimator for Self-Supervised Image Denoising
Yunhao Zou, Chenggang Yan, Ying Fu
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 patch variance loss for the noise estimator (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 24.21 -> 32.27 | Improved |
| decay patch variance loss weight to zero over training instead of keeping it constant (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 31.86 -> 32.27 | Improved |
| use patch size 8 for patch variance loss instead of global or other sizes (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 32.18 (global) -> 32.27 | Improved |
| add N2N loss ramped up over training (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 31.81 -> 32.27 | Improved |
| include blind-spot network for initial coarse denoising (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 14.46 -> 32.27 | Improved |
| anneal blind-spot loss weight to zero instead of keeping it constant (ablation) | DCD-Net · Kodak, Gaussian sigma=25 | PSNR (dB): 31.75 -> 32.27 | Improved |
| DCD-Net versus B2U on Poisson noise | DCD-Net · Kodak, Poisson lambda=30 | PSNR (dB): 31.64 (B2U) vs 31.60 (ours) | No clear effect |
| use DCD-Net instead of B2U for training cost | DCD-Net · SIDD raw-RGB | relative training cost: x17 -> x3 | Improved |
| R2R using known noise level prior on real noise | R2R · SIDD raw-RGB | PSNR (dB): 51.40 (ours) vs 47.20 (R2R) | Hurt |
Closest papers by meaning
- High-Quality Self-Supervised Deep Image Denoising (NeurIPS 2019 · similarity 0.92)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.92)
- Self-supervised Image Denoising with Downsampled Invariance Loss and Conditional Blind-Spot Network (ICCV 2023 · similarity 0.91)
- Self-Supervised Image Denoising Using Implicit Deep Denoiser Prior (AAAI 2023 · similarity 0.91)
- Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · similarity 0.90)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.89)
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.
