Research map / Label-efficient and robust learning / Noisy labels
Transfer Learning From Synthetic to Real-Noise Denoising With Adaptive Instance Normalization
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the AIN denoiser trained on synthetic noise without real-data fine-tuning | DND · DND | PSNR: 39.53 vs 39.23 | Helped |
| Transfer-learn selected parameters using real-noise data instead of training on synthetic noise alone | AINDNet · SIDD validation | PSNR: 35.66 -> 38.95 | Helped |
| Transfer-learn selected parameters rather than retraining all parameters | AINDNet pretrained on synthetic noise · DND | PSNR: 39.21 -> 39.37 | Helped |
| Apply geometric self-ensemble | AINDNet(S) · DND | PSNR: 39.53 -> 39.77 | Helped |
| Replace AIN-ResBlock with instance normalization and concatenated noisy image/noise map (ablation) | AINDNet(S) · DND | PSNR: 38.53 vs 39.53 | Hurt |
| Replace AIN-ResBlock with a standard Res-Block (ablation) | AINDNet(S) · SIDD validation | PSNR: 34.93 vs 35.19 | Hurt |
| Use the proposed multi-scale asymmetric estimation loss instead of L1 plus total-variation losses (ablation) | AINDNet(S) · SIDD validation | PSNR: 35.08 -> 35.19 | Helped |
| Use the proposed estimator and loss instead of FCN with asymmetric and total-variation losses (ablation) | AINDNet(S) · SIDD validation | PSNR: 34.90 -> 35.19 | Helped |
| Update AIN modules during transfer learning rather than freezing them (ablation) | AINDNet+TF · SIDD validation | PSNR: 34.60 -> 38.90 | Helped |
| Update the noise-level estimator during transfer learning rather than freezing it (ablation) | AINDNet+TF · SIDD validation | PSNR: 38.71 -> 38.90 | Helped |
| Update the last convolution during transfer learning rather than freezing it (ablation) | AINDNet+TF · SIDD validation | PSNR: 38.75 -> 38.90 | Helped |
| Transfer-learn with 32 real image pairs rather than training RIDNet with all its real data | AINDNet+TF · SIDD validation | PSNR: 38.71 -> 38.75 | Helped |
Closest papers by meaning
- Noisier2Noise: Learning to Denoise From Unpaired Noisy Data (CVPR 2020 · similarity 0.88)
- LAN: Learning to Adapt Noise for Image Denoising (CVPR 2024 · similarity 0.88)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.87)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.87)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.87)
- Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable (CVPR 2025 · 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.