Research map / Image generation / Image restoration
Dual Adversarial Network: Toward Real-world Noise Removal and Noise Generation
Zongsheng Yue, Qian Zhao, Lei Zhang, 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 noise generator to joint training (vs BaseD) (ablation) | DANet · SIDD validation (denoising) | PSNR: 39.19 -> 39.30 | Improved |
| add noise generator to joint training (vs BaseD) (ablation) | DANet · SIDD validation (denoising) | SSIM: 0.907 -> 0.916 | Improved |
| add denoiser to joint training (vs BaseG) (ablation) | DANet · SIDD validation (noise generation) | PGap: 4.07 -> 2.06 | Improved |
| add denoiser to joint training (vs BaseG) (ablation) | DANet · SIDD validation (noise generation) | AKLD: 0.223 -> 0.212 | Improved |
| retrain denoiser on generator-augmented data (DANet+) (ablation) | DANet · SIDD validation (denoising) | PSNR: 39.30 -> 39.47 | Improved |
| retrain denoiser on generator-augmented data (GDANet+) (ablation) | GDANet · Nam benchmark | PSNR: 39.91 -> 39.79 | Hurt |
| retrain denoiser on generator-augmented data (GDANet+) (ablation) | GDANet · DND benchmark | PSNR: 39.47 -> 39.58 | Improved |
| increase L1 weight tau1 from 1e+2 to 1e+3 (ablation) | DANet · SIDD validation (denoising) | PSNR: 38.66 -> 39.30 | Improved |
| increase L1 weight tau1 from 1e+3 to +inf (L1 only) (ablation) | DANet · SIDD validation (denoising) | PSNR: 39.30 -> 39.39 | No clear effect |
| add noise-statistics regularizer (tau2 0 -> 10) (ablation) | DANet · SIDD validation (noise generation) | PGap: 5.33 -> 2.06 | Improved |
| add noise-statistics regularizer (tau2 0 -> 10) (ablation) | DANet · SIDD validation (noise generation) | AKLD: 0.386 -> 0.212 | Improved |
| use only the regularizer without adversarial loss (tau2 +inf vs 10) (ablation) | DANet · SIDD validation (noise generation) | PGap: 2.06 -> 15.14 | Hurt |
Closest papers by meaning
- End-to-End Unpaired Image Denoising with Conditional Adversarial Networks (AAAI 2020 · similarity 0.92)
- GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images (ICLR 2021 · similarity 0.92)
- Image Blind Denoising With Generative Adversarial Network Based Noise Modeling (CVPR 2018 · similarity 0.92)
- Unsupervised Image Denoising in Real-World Scenarios via Self-Collaboration Parallel Generative Adversarial Branches (ICCV 2023 · similarity 0.92)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.90)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 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.
