Research map / Image generation / Image restoration
Learn From Unpaired Data for Image Restoration: A Variational Bayes Approach
Dihan Zheng, Xiaowen Zhang, Kaisheng Ma, Chenglong Bao
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 clean-to-noisy (C2N) instead of noisy-to-clean (N2C) data generation (ablation) | LUD-VAE · SIDD benchmark | PSNR: 34.74 -> 34.82 | Improved |
| add clean-to-noisy (C2N) instead of noisy-to-clean (N2C) data generation (ablation) | LUD-VAE · SIDD benchmark | SSIM: 0.901 -> 0.926 | Improved |
| add pre-processing Gaussian noise (sigma_x=15, sigma_y=10) versus none (ablation) | LUD-VAE · AIM19 | PSNR: 21.66 -> 22.32 | Improved |
| add pre-processing Gaussian noise (sigma_x=15, sigma_y=10) versus none (ablation) | LUD-VAE · AIM19 | MMD: 7.64 -> 2.34 | Improved |
| increase stochastic layers from 1 to 3 (ablation) | LUD-VAE · AIM19 | PSNR: 21.24 -> 22.32 | Improved |
| increase stochastic layers from 3 to 5 (ablation) | LUD-VAE · AIM19 | PSNR: 22.32 -> 22.07 | No clear effect |
| replace gamma-correction-only degradation with LUD-VAE learned degradation (ablation) | LLFlow · LOL | PSNR: 22.87 -> 24.92 | Improved |
| replace bicubic with nearest downsampling (ablation) | LUD-VAE · AIM19 | PSNR: 22.32 -> 22.65 | Improved |
| replace bicubic with bilinear downsampling (ablation) | LUD-VAE · AIM19 | PSNR: 22.32 -> 21.62 | Hurt |
| add LUD-VAE degradation instead of bicubic-only training data | ESRGAN · AIM19 | PSNR: 21.69 -> 22.32 | Improved |
| use LUD-VAE instead of DeFlow-NP (no pretrained network) | ESRGAN · NTIRE20 | SSIM: 0.6777 -> 0.7178 | Improved |
| use LUD-VAE instead of AWGN synthetic noise | DnCNN · SIDD benchmark | PSNR: 32.12 -> 34.82 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DeFlow — DeFlow: Learning Complex Image Degradations From Unpaired Data With Conditional Flows (CVPR 2021) · better on 3 of 6 shared results
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
- SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024 · similarity 0.90)
- Fully Unsupervised Diversity Denoising with Convolutional Variational Autoencoders (ICLR 2021 · similarity 0.87)
- Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · similarity 0.87)
- Tomographic Auto-Encoder: Unsupervised Bayesian Recovery of Corrupted Data (ICLR 2021 · similarity 0.87)
- DreamUHD: Frequency Enhanced Variational Autoencoder for Ultra-High-Definition Image Restoration (AAAI 2025 · similarity 0.86)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.85)
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.
