Research map / Image generation / Autoregressive and latent generation
Beyond Image Prior: Embedding Noise Prior into Latent Space of Conditional Denoising Transformer
Yuanfei Huang, Hua Huang
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 CondSA block with zero-vector prior (ablation) | Restormer · SIDD Validation | PSNR: 39.97 -> 39.96 | No clear effect |
| embed estimated noise prior by naive concatenation in CondSA (ablation) | Restormer · SIDD Validation | PSNR: 39.96 -> 40.09 | Improved |
| add linear fusion module (LFM) (ablation) | Condformer · SIDD Validation | PSNR: 40.09 -> 40.21 | Improved |
| use ground-truth noise prior instead of estimated (ablation) | Condformer · SIDD Validation | PSNR: 40.21 -> 40.23 | Improved |
| embed prior into Q, K and V instead of Q, K only (ablation) | Condformer · SIDD Validation | PSNR: 40.23 -> 40.19 | Hurt |
| add local smoothness criterion to patch sampling (ablation) | LoNPE algorithm · Urban100 | RMSE: 0.125 -> 0.021 | Improved |
| add local smoothness criterion to patch sampling (ablation) | LoNPE algorithm · SIDD-Medium | CV: 0.213 -> 0.048 | Improved |
| increase patch size from 16x16 to 32x32 (ablation) | LoNPE algorithm · Urban100 | RMSE: 0.020 -> 0.045 | Hurt |
| increase sampling ratio from 10% to 20% (ablation) | LoNPE algorithm · Urban100 | RMSE: 0.020 -> 0.031 | Hurt |
| condition denoiser on noise prior (Condformer vs Restormer re-trained on P-G) | Restormer · CBSD68 (sigma_s=0.3) | PSNR: 28.01 -> 28.14 | Improved |
| use Condformer vs Restormer on Gaussian noise (sigma_s=0) | Restormer trained on Gaussian · Kodak24 (sigma_s=0) | PSNR: 32.82 -> 32.81 | Hurt |
| condition denoiser on noise prior | Restormer · SIDD Validation | PSNR: 40.02 -> 40.21 | 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.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 5 of 7 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
- Convexity-Aware Noise Calibration: A Self-Supervised Framework for Noise-Level-Unknown Image Denoising (CVPR 2026 · similarity 0.90)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.90)
- Rethinking Deep Image Prior for Denoising (ICCV 2021 · similarity 0.89)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.89)
- Calibration-Free Raw Image Denoising via Fine-Grained Noise Estimation (TPAMI 2025 · similarity 0.89)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 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.
