Research map / Image generation / Image restoration
Reconstructing the Noise Variance Manifold for Image Denoising
Ioannis Marras, Grigorios G. Chrysos, Ioannis Alexiou, Gregory Slabaugh, Stefanos Zafeiriou
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 Rec subnet (noise-manifold autoencoder) to the generator (ablation) | Ours (empty c, ResNet) · DnD RAW | PSNR (dB): 47.51 -> 49.90 | Improved |
| add Rec subnet (noise-manifold autoencoder) to the generator (ablation) | Ours (empty c, ResNet) · DnD sRGB | PSNR (dB): 38.54 -> 41.50 | Improved |
| add residual learning on top of Rec subnet removal baseline (ablation) | Ours (empty c, ResNet) · DnD RAW | PSNR (dB): 46.92 -> 47.51 | Improved |
| add residual learning on top of Rec subnet removal baseline (ablation) | Ours (empty c, ResNet) · DnD sRGB | PSNR (dB): 37.73 -> 38.54 | Improved |
| use residual-domain discriminator instead of standard discriminator (ablation) | Ours (empty c, ResNet) · DnD RAW | PSNR (dB): 49.50 -> 49.90 | Improved |
| use residual-domain discriminator instead of standard discriminator (ablation) | Ours (empty c, ResNet) · DnD sRGB | PSNR (dB): 41.0 -> 41.50 | Improved |
| append the noise-manifold cGAN to UPI backbone | UPI · DnD RAW | PSNR (dB): 48.89 -> 50.05 | Improved |
| append the noise-manifold cGAN to UPI backbone | UPI · DnD sRGB | PSNR (dB): 40.35 -> 41.59 | Improved |
| append the noise-manifold cGAN to CBDNet backbone | CBDNet · Nam | PSNR (dB): 40.02 -> 41.20 | Improved |
| append the noise-manifold cGAN to CBDNet-JPEG backbone | CBDNet-JPEG · Nam (retrained with JPEG augmentation) | PSNR (dB): 41.31 -> 42.38 | Improved |
| append the noise-manifold cGAN to RIDNet backbone | RIDNet · SIDD | PSNR (dB): 38.71 -> 39.82 | Improved |
| give non-empty conditional info c (noise model, camera id) instead of empty c | Ours (ResNet) · DnD RAW | PSNR (dB): 49.90 -> 50.91 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RoCGAN — RoCGAN: Robust Conditional GAN (IJCV 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- End-to-End Unpaired Image Denoising with Conditional Adversarial Networks (AAAI 2020 · similarity 0.92)
- Dual Adversarial Network: Toward Real-world Noise Removal and Noise Generation (ECCV 2020 · similarity 0.91)
- GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images (ICLR 2021 · similarity 0.91)
- Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training (NeurIPS 2021 · similarity 0.90)
- Back to Basics: Let Denoising Generative Models Denoise (CVPR 2026 · similarity 0.90)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · 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.
